# History (/history)
## 2020 [#2020-step] * Selected as a finalist for the yearlong KL-YES Program US State Department exchange (fully funded). Selected in 2019, was supposed to start in August 2020. * KL-YES got cancelled due to COVID. They offered me a deferred spot for the following year but I declined it because I was 15, scared and completely unmoored by what followed. * My father left during the height of COVID. My mother moved us to my grandparents' house for stability. Faced a lot of financial insecurity, and for the next three years I felt like I had no ground beneath my feet and no roof over my head. * Out of sheer rage I did CS50, I loved the challenge because it was tougher than anything 8th grade offered and when the code finally compiled the satisfaction was unmatched. It was really tough as it's designed for Harvard and Yale undergrads. * I learned how to think algorithmically and break down problems into manageable pieces. I've applied this skill everywhere in my life. * It proved I could build things and create some ground even when my environment was collapsing. * Always used writing as my primary outlet for dealing with life.
## 2021 [#2021-step] * December 2021. Joined Dear Asian Youth (remote 501(c)(3)) as a Prose & Poetry Editor in the Literature department. DAY was remote-only and actively recruiting people * Started deeply engaging with the editorial process, learned a lot about empathetic reviews. * Respecting the writer's artistic integrity while giving critical feedback on what went good and what could be better. * You can fix bad writing but not an empty page. You edit toward the writer's intent and help them polish it. This is the core of workshopping with writers. This same logic shows up later in how I redesigned DIAL's recruitment where I filtered for potential instead of polish. * This was also the first time I observed a well-run remote organization up close with different roles, structured teams, Discord set up properly, real project management. I replicated this for Breezer's Brew and eventually the DAY engineering overhaul came from seeing what a functional version of this could look like. * Published in *Lamp Poetry Collection* ("Death, Alas", "Nostalgia").
## 2022 [#2022-step] * **Academics:** Completed IGCSEs. Received the Edexcel High Achievers Award and Daily Star High Achievers Award (given to students with 7+ A/A\*s in one sitting). * ALEX Program (American Leadership Exchange): Selected as 1 of 5 from Bangladesh as a YES finalist. Fully funded two-week State Dept. exchange in Washington D.C. and Iowa. Leadership training and community volunteering. * The Daily Star (Contributor): * Wrote opinion pieces and poetry. * Experienced the exploited creative reality with extremely low wages, toxic atmosphere, and editors changing (and publishing!) my work without consent. * I stopped going to the meetings and eventually quit. I knew what good editing feels like. * The Breezer's Brew (Founder/EIC): * Established South Breeze School's first literary club, scaling it to a structured organization with 50+ members, mentored writers and editors. * There was no platform for writing/art at school. I took what I learned from DAY and applied it here. I.e., stakeholder management, running a remote-only literary organisation at scale, proper project management, building mentorship and editorial guides culminating in the publication of the first volume. Remote-only was intentional to bypass school-time restrictions and so both SBS campuses could participate.
## 2023 [#2023-step] * Academics: Started AS in school. * A is the highest grade at the AS Level * Got 2 As and 2 Bs initially. Retook physics to turn the B into an A. * Ended up with 3 As and one B. * International Writing Program: Between The Lines * Full-scholarship literary residency (acceptance rate \<1% for internationals) by UIowa and US Dept. of State. (UNESCO City of Literature). Selected as 1 of 2 students from Bangladesh * Explored creative writing as a tool for cultural diplomacy. Online due to COVID restrictions. * First ever to be accepted from South Breeze School * They used a writing platform with optional peer review. I spent four hours giving individual written feedback to every single participant. By the end, all the instructors knew my name and it was the same instinct as the editorial work at DAY where I edited toward the writer's intent. * This shaped my brain in ways I'm still discovering because an art form I spent years freestyling and doing well in, was now revealed to me as serious pedagogy. The program director told me my writing had a lot of bilingual friction. That one comment has been in my mind ever since and has now shaped how I think about AI models and bilingual cognition. * The Breezer's Brew (Founder/EIC): * Published Volume 1 featuring artwork and writing from 14 students (Mentored around 30 I think, the other half didn't get published, got 200+ applications) * Distributed across local bookstores, got an ISBN, published pdf online * Baatighar, The Bookworm. I have photos * Magazine printing costs were too high and sales wouldn't break even. Bangladesh has strong garments manufacturing, we then designed and sold letterman jackets to fundraise $1150. Got so popular it became part of the uniform. * Generated BDT 140k here. Ended up with a 20k surplus that we re-invested into training courses * This was the first time I made so much money all at once. It really clicked that my life is in my hands and I didn't have to succumb to the financial instability. * International Publications: * Published in *The Paper Crane* ("my father, drunk, talking to me in bed"). (Indie mag) * Written in 2022, published in 2023.
## 2024 [#2024-step] * A family health emergency right before my A level exams and graduation meant hospitalisation and months of crisis at home (February - April) * Didn't know how we would survive without an active earner * When the most critical & foundational systems failed, systems that were supposed to protect me, I started obsessively working on architecting in long-term resilience and fault-tolerant structures, both human and technical. * This is a major reason my A2 grades took a hit that year * My return to tech started with a few key events * Made a website for The Breezer's Brew * My physics teacher offered me a web dev gig because he loved my website * First ever money I made that wasn't from writing and this paid a lot more and I loved the game. * Decided to learn web dev so I could take up freelance gigs. * A friend lied about having a $120k/year remote cybersecurity job and asked if I wanted to work on a startup. * I realised I did CS50 4 years back, and didn't really focus on it. I was all into writing. I remembered I genuinely enjoyed solving problems with tech and wanted to combine it with humanities. * Cofounded Sirkoi, designed the MVP (UI states, customer journey, website, logo). Left over a product-market-fit disagreement — the data didn't support the direction, and I wasn't willing to keep building around it. * Rate my professor clone pivoted to AI tutor matching in Dhaka * **Note: this was before the massive influx of vibecoded AI slop apps and sites started to flood everywhere. Everything was handcoded and required me to deeply learn and understand how it works.** * Joined DIAL in late 2024 * I learned about Stanford's Symbolic Systems where they combine CS, linguistics, philosophy and psychology. Realised this is the intersection I love being in. Discovered DIAL and HCI as the closest natural bridge to my interests of humane technology and social impact * Initially wanted help with an idea for a research project, scrapped due to lack of potential. This was more of an anthropology/effect-of-social-media direction where I wanted to see how certain groups of people could develop a completely different (western, individualistic, etc) mindset despite living in Bangladesh with my hypothesis being that we grew up on the internet and that shaped us heavily. Good question, bad research idea in hindsight. * Observed a disconnect between the research values of inclusivity and the lab's operational reality. The lab was mostly run by grad students and undergrads who had poor project management and communication skills. No proper mentorship and mostly running on informal communication. * I wanted there to be a proper pathway into doing research so I made one myself. Pitched an idea for an HCI internship for high school students in Bangladesh, the PI accepted the pitch but said DIAL's under capacity to handle an influx * Proposed software changes to make things easier and an onboarding guide to the teams working. She liked the idea but said they're stubborn so we could try it with newer projects. * Was used to qualitative research from IAL Psychology * Started working on DIAL's colearning project but it got paused * Worked on the Nonbinary NLP proposal while teaching myself how lit reviews and NLP methods worked * Wanted to make a more robust version of the MISGENDERED Hossain et. al (2023) paper. Benchmark for AI gender bias. * Founded the Engineering department @ Dear Asian Youth * DAY evolved from a student group and their problems were with scaling up. The ad-hoc systems they used lacked structure and didn't work at scale. Actively worked with different departments to understand their workflows and needs. * Led the migration to Payload CMS and the "Documentation as Code" initiative. * Implemented a design system, user journey map, redesigned literature workflow, implemented opensource systems for HR and Literature (Ghost & Frappe HR) * In both cases I realised I had to build foundational infrastructure for the organisations that needed to run better.
## 2025 [#2025-step] * Got the idea for Fieldkit (opensource nonprofit stack) after implementing a few systems for DAY * Solving Literature's Ghost usage problems proved that Standard Operating Procedures are infrastructure * Organically evolved with the opencore business model idea, pitched it and they loved it * Instead of rushing the HCI Internship, I chose to build DIAL's digital and operational foundations first * I wanted to build a workplace I would actually enjoy working in long-term. This was preventative maintenance for a system I intended to stay in. * Was actively researching for the website and built v1 of DIAL's recruitment and onboarding and pitched it. She liked it but I wasn't completely happy with it * May to October * Horrible Chikungunya & dengue with subsequent depression * Complete collapse, couldn't get anything done * Auctus was a T20 admissions counseling service that offered a 100% scholarship * Before I applied, I took their wall-of-words wordpress site, redesigned their entire platform while understanding their target audience to try and demonstrate value for the Feynman Fellowship, couldn't afford their BDT 500k fee otherwise * Got rejected. Said my ECs were one of the most unique and impressive he's seen but there's no way I would get a full ride given where my grades stood. * Focused fully on taking mocks at school and getting my grades back up * Took mocks, got all A\* predictions * Got $5k for Fieldkit from AWS Activate via DAY/Hack+ connection * Met with AWS Solutions Architects (Nov) to validate the "Nonprofit OS" architecture. * At DIAL * they asked for a website, I started building a research information management system that builds a website * their old screening made applicants present assigned papers which filtered for regurgitating academia. I redesigned it into a two-phase system with questions that detect curiosity and grit, then an onboarding that calibrates how each student gets managed.
## 2026 [#2026-step] * Preparing to retake A2 exams in June to finalize the academic comeback. * Did gig work to self-fund the 130k BDT registration fees due in February. Built the ground while standing on it. * Web dev freelance. * Planning to launch Autana Systems as a business. Was thinking of doing two things * AI-readiness consulting for SMBs * Converting ancient wordpress websites to be LLM-ingestable. This is a boring but stable Agentic SEO business. * Releasing opensource tools for researchers * Doing automation work to make their organisation legible to AI
# Preface (/preface) # Hello World [#hello-world] I'm Uzayer Masud, I'm 21 and I used to be, above everything else, a writer. I started when I was 6 years old and primarily did just that up until 2024. I worked as an editor for years, have multiple published pieces that I'm still somewhat fond of, a magazine I built from scratch by mentoring writers, and countless unpublished pieces in my writing folder. Since then, I've mostly done (and implemented) system design at the sociotechnical and organisational layer. My foray into tech and research was very non-linear and interdisciplinary and I am still finding my way through it. At every step of the way, I wanted to create something, and the infrastructure I needed didn't exist, so I built it for myself and those who come after me. I believe in resilient systems that outlast individual actors and are designed for humans in mind. Most of this work is structural (therefore difficult to quantify and demonstrate) but its effects and benefits compound later at scale. My dream was to study Stanford's Symbolic Systems (intersection of CS, Linguistics, Philosophy and Psychology), now I'm planning to do research and apply for a PhD in my final year of undergrad. My interests are in linguistics and cognition, NLP, AI ethics, cognitive science-inspired AI models, human-computer-interaction and system design. I cannot, for the life of me, let go of the explanations of the "why this matters" and "here's the thought process and how everything connects." I almost always think in graph-structures and have a hard time linearising my thoughts, hence the metacognitive philosophy-filled mess this doc has turned out to be. I'm also somewhat good at business. ## Colophon [#colophon] ### A *colophon* is a snobby designer word for 'how this doc was made' [#a-colophon-is-a-snobby-designer-word-for-how-this-doc-was-made] I started writing this in November 2025, when after a long depressive period I realised I could macrothink in the projects I did but struggled to connect the dots and find proper semantic throughlines in the work. I've revisisted this doc countless times, both as trying to build the building blocks for my college applications at the time, but also as a reminder to myself that I exist and the work is real. The [History](/history) section contextualises my projects in what happened to me in my (so far short) life and expains the decisions I made. The [projects](/projects) are structured with a What, Why and Additional Info. Everything else is reference material. I have some publications linked at the end if you have the appetite to read more of my writing. This document might be the densest piece of work I've produced so far. # Reference (/reference) ## Experience [#experience] Roles held across and .
### Literature Editor [#literature-editor-step] *Dec 2021 – Aug 2024 · 2 yrs 9 mos* * Regularly researched topics on current affairs, humanitarian issues, and marginalised communities * Edited articles for clarity; cross-referenced sources to verify information * Collaborated with other editors on poetry and prose - fiction, personal narratives, journalistic articles, opinion pieces
### Website Team [#website-team-step] *Jul 2023 – Aug 2024 · 1 yr 2 mos* * Managed administrative tasks for DAY's website on Wix, ensuring content updates and structural improvements * Assisted with website maintenance, organisation, and minor updates to improve user experience * Supported the team in maintaining a functional and accessible digital presence aligned with DAY's advocacy mission
### Head of Engineering [#head-of-engineering-step] *Aug 2024 – Present · 1 yr 5 mos* Leading DAY's engineering and technical infrastructure. See [Organisational Systems](/projects#organisational-systems) for the projects built in this role.
### Research Assistant [#research-assistant-step] *2024 – Present · 1+ yr* Working with Dr. Nova Ahmed at [DIAL (North South University)](https://dial.northsouth.edu) - an HCI lab in the ECE department. * First independent project: [*MISGENDERED++*](/projects/misgendered-nlp), a robustness evaluation framework proposal for nonbinary pronoun bias in LLMs (archived when a concurrent paper solved the underlying problem) * [Redesigned DIAL's RA recruitment and onboarding pipeline](/projects/dial-recruitment) - PI accepted the redesign; developing the post-recruitment layer * Building [LabKit](/projects/labkit), a Research Knowledge Orchestration Platform - a semantic graph-based alternative to standard CMS tools for academic labs (PayloadCMS 3 + Next.js + MongoDB) * Exploring [ResearchNet](/projects/researchnet): grant compliance automation on top of LabKit's data model * Pitched an [HCI research internship for high school students in Bangladesh](/projects/hci-internship) See [Research & Lab](/projects#research--lab) for full project detail. ***
## Programmes [#programmes] *University of Iowa & U.S. Department of State · 2023* Two-week fully funded literary residency. 1 of 2 Bangladeshis selected; 1 of 2 students selected nationwide by U.S. Embassy Dhaka (acceptance rate \<1%). Studied at IWP - UNESCO City of Literature. Explored creative writing as a tool for cultural diplomacy. *U.S. Department of State, KL-YES, American Councils · 2022* Two-week student exchange program. One of 5 Bangladeshis selected. Leadership training for multicultural teams; volunteering at Capitol Area Food Bank and Living Lands & Waters. *** ## Publications [#publications] *International Writing Program, University of Iowa* · Essay "Nazrul's Immaculate Rizz - My Experience with Bangla and Internet Culture" *Published anthology* · Poetry Two poems: *Death, Alas* (p. 99) and *Nostalgia* (p. 164). [Goodreads](https://www.goodreads.com/en/book/show/56953476-lamp-poetry-collection) · [Excerpt →](https://drive.google.com/file/d/1wTV_6apjrlpqoVV3GAkBYmUD6foqw8gq/view?usp=sharing) *Published anthology* · Poetry *my father, drunk, talking to me in bed* (p. 80). [Amazon](https://www.amazon.com/dp/B0CV6SSMYP) · [Excerpt →](https://drive.google.com/file/d/1PZMDF1JHB9s3lN4iVQFT_Dt0SD1onUDA/view?usp=sharing) *The Daily Star* · Fiction Published in the SHOUT Fable Factory section. *The Daily Star* · Essay *The Daily Star* · Fiction Published in the SHOUT Fable Factory section. *The Daily Star* · Essay Kind of embarrassed by this one. Was very young, dumb, and in love back then. *The Daily Star* · Performance Performed twice: [Jan 24, 2024](https://www.instagram.com/reel/C2j6RUGpzCF/?utm_source=ig_web_copy_link\&igsh=MzRlODBiNWFlZA==) and May 7, 2025 (no recording available). *South Breeze School* · Mixed Published two original pieces in the inaugural issue. The magazine was sold at Baatighar and The Bookworm. Raised BDT 140k (\~$1,150) via letterman jackets to cover printing costs. *** ## Press [#press] > "Another writer, Uzayer Masud, says 'poetry is where we start to push the bounds of language to describe things that we don't have words for. For example, consider how you might describe the colour red to someone who has never seen it before. There are no words in \[any] language that can describe red for what it is.'" * [Into The Rhyme and Reason of Poetry](https://www.thedailystar.net/daily-star-books/news/the-rhyme-and-reason-poetry-3277106), *The Daily Star* *** ## Organisations [#organisations] A key to the organisations that appear throughout this site. **Dear Asian Youth (DAY)** - 501(c)(3) advocacy nonprofit. Founded 2020 by Stephanie Hu (Harvard). [dearasianyouth.org](https://dearasianyouth.org) **DIAL, North South University** - HCI lab in the ECE department of NSU. Lab director: Dr. Nova Ahmed. Founded 2020. [dial.northsouth.edu](https://dial.northsouth.edu) **The Daily Star (TDS)** - Largest English newspaper in Bangladesh. Youth segment formerly SHOUT, now rebranded to Campus. **[Breezer's Brew (TBB)](/projects/breezers-brew)** - South Breeze School's first student-led literary magazine. **International Writing Program (IWP)** - University of Iowa IWP. UNESCO City of Literature. Hosted the Between The Lines residency programme. **Autana Systems** - A company I'm planning to start. If documentation systems are integral to the success of enterprise AI agents, what can Autana do in the space of documentation-as-a-service? GitLab's handbook had a small effort to create a company handbook template - now largely abandoned. Even if it's not git-and-markdown based, how far can AI go in generating guardrails and documentation for itself? Where does the real challenge lie in translating messy business logic into documentation that an agent can actually follow? # Components (/test) ## Code Block [#code-block] ```js console.log('Hello World'); ``` ## Cards [#cards] # Auctus Admissions Website Redesign (/projects/auctus) *2025 - Archived* ## What [#what] * There are a couple T20 consulting agencies in Bangladesh but they need to differentiate themselves from the pay-to-play agencies that send kids to the UK, Australia, Malaysia, etc * They usually charge a lot of money (justified but I can't afford it) and Auctus was the only one with a program where if you got accepted to their program they wouldn't charge you anything * I did a complete visual and operational redesign of the *Auctus Admission Consulting* website, transitioning it from a generic WordPress-style site to a modern, value-driven interface. * Built a custom Admissions Value Proposition framework to differentiate them from "pay-to-play" agencies in Bangladesh. * They ended up price gouging me for free labour. Three months later they sent an email asking to meet and if I'd build their site. I replied within hours and the next day, even before giving me a meeting, they told me they moved on with another developer. Another month later, I saw the site they made and it was a copy of the draft I showed them. Plagiarised my work completely. ## Why [#why] * Built and delivered the full product *before* applying for their **Feynman Fellowship** to demonstrate initiative hoping they'd accept me. * The founder confirmed the site was great, but rejected the application due to where my grades stood (May 2024 session) * He said my ECs were impressive but there's no way I'd get a full ride because of my A2 grades. Said the call was a courtesy since I put so much effort into building something for them. * Also said they get 700+ applications for a cohort of 20 i.e Bangladesh is overpopulated and I solved a problem that didn't need solving * This failure directly catalysed my A\* Predicted Grades in late 2025 Can re-deploy the site I made if it helps - currently down. # Breezer's Brew (/projects/breezers-brew) *2022 – 2023* · ## What [#what] * South Breeze School's first literary organisation with 50+ members * Mentored writers, editors and artists to create a complete indie mag * Published + distributed to bookstores + ISBN registration * Available as a PDF online * Took inspiration from DAY's literature department on how to run it, effectively ran the same model with tweaks for the Bangladeshi context * Learned InDesign and editorial design * Raised BDT 140k+ (\~ $1150) by selling letterman jackets - a response to the magazine not raising enough to break even * They loved it so much the jackets became part of the school uniform. They want it every year now ## Why [#why] * I wanted a platform to showcase literary talent and for writers to hone their craft * Didn't have a single platform available so I built my own. ## Additional Info [#additional-info] * The breezersbrew\.org website is currently down, it has the full pdf + more publications * Honestly 2023 feels so long ago I forgot what I was doing but this took me over a year of prep and fundraising ## Related [#related] * [DAY Design System & Marketing Funnel](/projects/day-design-system) — Breezer's Brew gave me hands-on experience running a literary organisation; that knowledge fed directly into how I structured DAY's literature infrastructure years later # DAY Design System & Marketing Funnel (/projects/day-design-system) *2024 – Ongoing* ## What [#what] * Led a UI/UX team to audit DAY's design history, then create a new visual direction that reflected how DAY grew as an organisation. Developed a design system based on `shadcn/ui` and Tailwind CSS to standardise the visual language. * Collaborated with Marketing and senior partners to create a user journey map that models the lifecycle of the different archetypes/personas of people that interact with DAY to their ultimate goals so we could identify pain points at each stage and mitigate them. ## Why [#why] * Brand consistency was impossible to enforce manually due to the sheer size of DAY. We needed a component-based system where "correct" design is the default state, ensuring a professional appearance regardless of who is building the page. * The organization lacked a clear understanding of why engagement was low. Mapping the customer journey allowed us to treat recruitment and engagement as a systematic funnel to identify specific friction points in the user experience. Currently being completely overhauled. ## Related [#related] * [Fieldkit OSS](/projects/fieldkit) — this design system is one of Fieldkit's core modules, being validated at DAY before extraction into a template * [Literature Workflow Redesign with Ghost CMS](/projects/ghost-cms-workflow) — the content system that surfaces through this design language * [GitLab-Inspired Handbook](/projects/gitlab-handbook) — the knowledge infrastructure this design system governs # DIAL Recruitment & Onboarding Redesign (/projects/dial-recruitment) *Late 2024 – Ongoing* Status: PI accepted the redesign. Doing one more pass to develop the post-recruitment layer. What happens after recruitment is the harder problem. ## Why [#why] * DIAL's legacy recruitment system is a funnel: * Cohort taken per semester * Assigned three HCI papers from the lab, they have to synthesise then present the papers (slideshow) * Finish three ethics modules from TREE (what an IRB is, working with human subjects, do no harm) *Figure 1: Legacy Recruitment Pipeline* * I had to go through the same screening process. I made all three presentations so I could start working immediately. Halfway through the second presentation the lab staff stopped me and said I don't have to do the rest. He said that my slide designs were "sober" and my English was good * It felt weird to get complimented on those criteria when I expected more depth on the research. I won the screening here because I'm heavily bilingual and think in both languages so presenting a paper and speaking in English wasn't a problem for me. However I believe that shouldn't be the criteria that I should win on. * DIAL's recruitment pipeline completely contradicted their values of inclusivity as it's based on an NCTB-inspired rote memorisation pedagogical approach. Hinders from developing talent. * NSU's incoming freshman class has a sizable majority that is unfamiliar with making slides let alone presenting them. Testing them on an ability they haven't been taught yet is unfair. Testing them on reading comprehension when they don't know how to read an academic article in the first place is also very unfair. This leads to memorisation, AI plagiarism and reading text off the slides. * **If we are to be inclusionary, we must look for reasons to admit applicants who might not be polished, but have potential. We needed a system that could detect raw signal amidst the noise of an unpolished freshman.** * I went through so many versions of this while designing it because I realised **my design would fail in NSU's context because my frame of reference was wrong**. Originally, I felt unfairly advantaged so I redesigned it to have two phases. That was in May 2025. This time around I felt like there needed to be more guardrails and notes for each question (what signal I'm trying to extrapolate, observations & actionable suggestions). I ended up building out all of it to realise I hadn't considered my positionality. * The world I'm in is AP & IB & IAL -> Ivy League/YCombinator. Everyone's hypercompetent, collaborative and hungry to do more and do it better. I was filtering for top talent when I needed to filter for unpolished potential. * I started again from scratch, made a signal dictionary of what I want to measure. The questions are a frontend for the signals (grit, resilience, creativity, etc) and the questions can be changed. The underlying structure is the signals are then split into phases: pre-screening, secondary screening, onboarding, recurring 1:1s. ## What [#what] A complete overhaul of DIAL's recruitment pipeline to adapt for the agentic era and test students on their curiosity and grit and qualities that are hard to fake. Designed for applicants with constraint-laden backgrounds in mind. *Figure 2: Redesigned Recruitment Pipeline* * **Redesigned System:** A progressive resolution system split into three phases: 1. **Screening:** A diagnostic questionnaire testing for values-fit, curiosity, and observation skills rather than academic jargon. 2. **Simulation:** They still present a paper with slides as the format, as that's what the lab is used to, but we're no longer evaluating whether they understood the mechanics of the paper. The question is whether they can connect what the paper argues to something they've personally observed or experienced. A student who grounds a finding in their own neighbourhood is showing more than one who recites the methodology. 3. **Calibration:** A standardised onboarding protocol that configures the management style (incentives, scaffolding, risk) based on the student's specific profile. *Figure 3: Theory of Change - Progressive Resolution System* * **The Signal Dictionary (Backend):** Capturing "messy" variables like curiosity by grounding them in high-fidelity observations (e.g., identifying friction in the design of daily objects). * This is an informal vibes-based codebook trying to codify observable behaviour. This redesign comes from my experience leading teams and observations while at DIAL, there's no dataset I can pull to validate what makes a good research assistant but I can rely on inter-rater reliability as a blunt heuristic. * The logic maps onto the same problem where if different reviewers are assessing curiosity independently then you have to define what curiosity actually looks like in the room. * **Hard-coded Empathy:** The standardised rubric embeds inclusive values directly into the review process so that correct behaviour doesn't depend on any individual's judgment call. Bolted on inter-rater reliability. * Replaced TREEE (a biomedical ethics module) with Macquarie University's social science ethics course and the Center for Humane Technology's course as the old training was contextually irrelevant to HCI fieldwork. **This system only works because of the qualitative component to DIAL's research.** My view is that this is a systems design problem. The PI wants and practices inclusive recruitment, but that value isn't encoded into the structure - it ends up being a preference without an enforcement mechanism. I redesigned to apply software patterns to fix that: define the signals, build sensors, configure outputs. DIAL often works for low-resource contexts, this reframes it from "charity work" to a practical engineering constraint that's applicable everywhere. This redesign treats the lab as a complex social system where technical abstractions are used to model and preserve human empathy against systemic entropy. ## Additional Info [#additional-info] * As of today, with the old process, you can throw the entire paper's pdf into Claude for PowerPoint and have it output a well-designed slide that you can just read off of. The entire playing field has changed. * I wrote these questions to gauge how much they actually engage with the research and relate to their own personal experiences. Even if they use ChatGPT to answer them, they will have still learned something in the process and that's still a win. * There might be a lot of kids at NSU that don't really care about problems in Bangladesh. They're surviving through here and their goal is to go abroad for a masters. They might think that working for low resource contexts might be beneath them because they're trying to solve a local problem. They're the harbingers of funding so they'll work on high resource problems instead. * They need to understand that in the context of software, efficiency is not a bad thing. When vertical scaling is free then the apps are ram-hungry slop machines. When forced into resource constraints they are forced to improve and lower long term costs too. ## Related [#related] * [LabKit](/projects/labkit) — DIAL's research platform; solving capacity constraints is the other half of the recruitment problem * [Nonbinary NLP Research Proposal (MISGENDERED++)](/projects/misgendered-nlp) — my first project at DIAL, which gave me the context to redesign recruitment * [HCI Internship for HS Students](/projects/hci-internship) — the downstream program that depends on this redesign being in place first # Fieldkit OSS (/projects/fieldkit) *2025 – Ongoing* ## What [#what] * An opinionated and reproducible tech stack for nonprofits and student organisations designed with lean, remote-first and async-by-default practices in mind. Includes a template for an organisational handbook and to make everything LLM-ingestable by default. The core of it is designed to run on low-cost infra with minimal maintenance. * Newer organisations do not carry legacy practices or have multiple people to train. This lays concrete foundations they can build on top of, keeping in mind that everything will be legible to an agentic intelligence layer. * In my experience I've seen a few patterns emerge organically * Student organisations start out with smaller teams, they move fast, they ignore some standard practices like maintaining documentation, doing project management or maintaining communication. This is only possible in teams that are small and in-person. The bad practices balloon into organisational debt that cannot be handled when scaled up. * The debt does take a toll when onboarding is oral tradition and all information exists in people's heads or in lost messages and emails. * Technical note: this is *not* building software from scratch, the project is all containerised applications arranged in an opinionated way. It is far more feasible to maintain the glue that holds it together than building the software from scratch up. * **Examples** * [Frappe HR](https://frappe.io/hr) is an excellent opensource answer to expensive and heavy HR systems. * [Plane](http://plane.so/) took inspiration from [Linear](https://linear.app/) and has the same underlying mental models that led to Linear's success except they adapted it to be more flexible. * Usually project management tools target enterprise software teams, are too complicated, opinionated and expensive. * [Plausible](https://plausible.io/) is a very lightweight and privacy focused Google Analytics alternative. * Using [Authentik](https://goauthentik.io/) for a unified dashboard experience **Figure 1:** The Fieldkit stack. Authentik provides a unified SSO layer across all services, grouped by function. ## Why [#why] * Giving people the tools without an underlying structure will likely end up in abandonment. A core part of Fieldkit is the standard organisation handbook that should be maintained as operational docs for teams involved. It's minimal by design and is meant to be adapted by each organisation. * At a mid-sized organisation, SOPs are infrastructure. You cannot scale software without scaling culture. I treat the organisation's handbook as **"wetware infrastructure"** - like code that runs on humans and agents. Fieldkit deploys both the software (Docker containers) and the culture (Markdown Handbooks) in a single "install" that organisations can then adapt to their needs. * I've observed that the conditions for deploying effective AI agents (high traceability, zero data silos, rigorous documentation) are identical to the conditions for a healthy, remote-first workplace. Fieldkit enables this organisational health, but the system will degrade if the human culture does not maintain these standards. * We enforce a "Standard Configuration" because maintaining the "glue" for every possible software combination is technically impossible for us and cognitively exhausting for the user. We take on the decision debt so they don't have to. * There is a market failure for mid-sized impact organisations. They are served either by free consumer tools (which don't scale) or enterprise SaaS (which they can't afford). There are some great open-source tools that solve those individual problems * This is similar to opinionated Linux distros like Omarchy. Similar to how senior dev scaffold projects. * I'm trying to lower the cognitive tax of creating an organisation and hitting the ground running. * Most of these organisations will not scale beyond 30-50 people, especially if it's a high-churn environment like student organisations or research labs. If it's a company, this setup still avoids vendor lock-in because scaling up means vertically scaling to a bigger VPS (still a marginal cost increase) or migrating data to other tools which they all support. ## Additional Info [#additional-info] Received **$5k in AWS Activate Credits** via DAY/Hack+ and had initial conversations with AWS Solutions Architects (Nov 2025) about the architecture. * DAY is the active proving ground. Each DAY project is a Fieldkit module being validated in a real org: * **Ghost CMS** literature workflow (live) * **Frappe HR** self-hosted HR system (live) * **GitLab-style handbook** Astro + Starlight, docs-as-code (live) * **Design system + user journey mapping** shadcn/ui + Tailwind (live) * **Next.js + Payload CMS** site rebuild (in progress) * Currently mid-restructuring at DAY with a consultant. The outcome of this overhaul becomes the Fieldkit template. The goal is to extract a reproducible, opinionated stack from a real deployment rather than designing for a hypothetical org. * If the DAY model holds, Hack+ is a natural distribution channel. They already provision new orgs with technical infrastructure via vendor partnerships, but most orgs lack the expertise to use it. Fieldkit bridges that gap as an open-source project. ## Related [#related] * [Literature Workflow Redesign with Ghost CMS](/projects/ghost-cms-workflow) — the Ghost module, validated at DAY's literature department * [GitLab-Inspired Handbook](/projects/gitlab-handbook) — the docs-as-code module, running live at DAY * [DAY Design System & Marketing Funnel](/projects/day-design-system) — the design system module, also live at DAY * [LabKit](/projects/labkit) — same philosophy applied to research labs; both are opinionated, reproducible infra stacks for specific org archetypes # Literature Workflow Redesign with Ghost CMS (/projects/ghost-cms-workflow) *2024* ## What [#what] * DAY Literature is the largest department construed of smaller teams with lead editors and writers, a separate publishing team works on getting things live. The workflow is complex and lives inside a multi-tab Google spreadsheet that serves as team management, reference library and project tracking. * The previous workflow had a big human error surface area due to gruntwork repetition. Applied software DRY principles to migrate to a simpler open-source Ghost CMS & Google Docs architecture. * Redesigned the architecture to treat Editors as first-class citizens, providing them with dashboards, credit, and better tools. * The legacy workflow (Google Sheets) treated articles as static files hidden in tabs. The new workflow (Ghost Dashboard) treats articles as state objects (Pitch -> Review -> Published). This shift from file management to state management reduces the cognitive load on the team, as the status of every piece of content is visible at a glance. * To solve the invisible editor problem, I utilised Ghost's tag system (e.g., tagging a post `Edited by: Chloe Dong`). This forces the CMS to auto-generate a public portfolio page for every editor so they have a shareable record of their contributions. * Credit allocation is structurally enforced by design instead of being left as an afterthought. * Decoupled the act of writing (Google Docs or whatever the writer wants) from the act of publishing (Ghost). This allows writers to work in a familiar environment without wrestling with a CMS, while editors control the final rendered state. ## Why [#why] * The previous workflow only made sense if you were in the team for a long while. Simplifying it was proper engineering in my opinion because then it's less prone to breaking. Maintenance tax is real sadly and the previous one already had lapses in records. * Editors got credit at the end of the article, with their names written in different formats sometimes. The data design had no space for editors as first-class entities next to writers. * Moving to Ghost (Open Source) aligns our infrastructure with our values of transparency and ensures we own our data rather than renting it from a closed garden like Wix. ## Additional Info [#additional-info] * Used webhooks to listen for tag changes. When an article is tagged `#editorial-review`, the system automatically creates a Google Doc from the pitch, links it back to Ghost, and alerts the editing team on Discord. This removes the manual friction of setting up documents. * Visibility is the currency of contribution. Using the tagging system to generate editor portfolios was my way of re-engineering the symbolic economy of the team. ## Related [#related] * [Fieldkit OSS](/projects/fieldkit) — this workflow is one of Fieldkit's core modules * [DAY Design System & Marketing Funnel](/projects/day-design-system) — the design language this CMS surfaces content through * [GitLab-Inspired Handbook](/projects/gitlab-handbook) — the other DAY infrastructure project running alongside this # GitLab-Inspired Handbook (/projects/gitlab-handbook) *2024 – Ongoing* ## What [#what] * DAY is completely remote and async by default. This type of organisation won twice, once during COVID, again during the agentic era. **This means the entire organisation needs to be legible to both remote workers and AI.** * A public, open-source, version-controlled handbook (built with Astro + Starlight) that serves as the single source of truth (SSoT) for the entire organisation. * An org handbook exists to answer the question of how and why something is done. It is the company knowledge distilled onto written form. This is an intentional practice that is incredibly annoying to do at first but pays compounding dividends the longer you do it. Sections have DRIs (directly responsible individuals) and anybody can contribute. * This is documentation as code. Changes are submitted as pull requests reviewed by code owners, and versioned in Git. * A shift from information siloed in heads or DMs to explicit knowledge (live at `handbook.dearasianyouth.org`) ## Why [#why] * DAY has 300+ remote volunteers, information needs to flow fast and free to the person that needs it. This is both async and scaleable. * **If it's not written down, it doesn't exist** * We adopted GitLab's philosophy to have a handbook that ensures that every member operates on the same shared reality. * Given the structure, this forces the person to contextualise their change in the larger scheme of how the organisation works. They have to locate the section where their change would be, and see what else gets affected. This is something AI is still notoriously bad at. * Anyone in the organisation, or even outside it, can submit a change. This is to **decentralise governance** and take a flatter and more community oriented approach. The DRI only needs to see the change, and then accept/deny it. * The pull request workflow is intentional friction that functions as a consensus algorithm. It forces ambiguity to be resolved *before* a policy is merged. Unlike a Google Doc where edits are fluid and often ignored, a Git-based workflow requires explicit approval from stakeholders and creates a rigorous audit trail of decision-making. * Institutional Memory as distributed ledger * We tried to reframe management as an information retrieval problem. High-turnover organisations suffer from knowledge decay. The handbook acts as externalized long-term memory that preserves the "state" of the organisation independent of the specific humans currently staffing it. It transforms the org from a volatile to a persistent system. ## Additional Info [#additional-info] * **Restructuring & Fieldkit Integration:** As of April 2026, DAY is undergoing a massive organisational restructuring. This handbook is the live production environment for this change. We are actively rewriting all SOPs that hold the new structure together, directly applying the wetware code philosophy from [Fieldkit OSS](/projects/fieldkit). * **Architecture:** Directories map to departments and `.codeowners` files assign maintenance responsibility to specific leads, ensuring documentation never goes stale. At GitLab our handbook is extensive and keeping it relevant is an important part of everyone's job. It is a vital part of who we are and how we communicate. We established these processes because we saw these benefits: 1. Reading is much faster than listening. 2. Reading is async, you don't have to interrupt someone or wait for them to become available. 3. Talent Acquisition is easier if people can see what we stand for and how we operate. 4. Retention is better if people know what they are getting into before they join. 5. On-boarding is easier if you can find all relevant information spelled out. 6. Teamwork is easier if you can read how other parts of the company work. 7. Discussing changes is easier if you can read what the current process is. 8. Communicating change is easier if you can just point to the diff. 9. Everyone can contribute to it by proposing a change via a merge request. One common concern newcomers to the handbook express is that the strict documentation makes the company more rigid. In fact, writing down our current process in the handbook has the effect of empowering contributors to propose change. As a result, this handbook is far from rigid. You only need to look at the [handbook commits list](https://gitlab.com/gitlab-com/content-sites/handbook/-/commits/main) to see the evidence. Every attempt is made to document guidelines and processes in the handbook. However, it is not possible to document every possible situation or scenario that could potentially occur. Just because something is not yet in the handbook does not mean that it is allowed. GitLab will review each team member's concern or situation based on local laws to determine the best outcome and then update the handbook accordingly. If you have questions, please discuss with your manager or contact the [People Success](https://handbook.gitlab.com/handbook/people-group/) team. ## Related [#related] * [Fieldkit OSS](/projects/fieldkit) — this handbook is a Fieldkit module; the pattern being extracted into a portable template * [DAY Design System & Marketing Funnel](/projects/day-design-system) — the design layer that this knowledge infrastructure pairs with * [Literature Workflow Redesign with Ghost CMS](/projects/ghost-cms-workflow) — the other core DAY operational system built around the same time # HCI Internship for HS Students (/projects/hci-internship) This is still extremely rough. Blocked on external factors - specifically [LabKit](/projects/labkit) reaching a stable state first. *Pitched Late 2024 – Planned* ## What [#what] * An HCI research internship for high school students in Bangladesh * Bangladesh has technical implementation heavy ECs for HS students like olympiads and competitions but not research programs because they're slower and heavier * CHRF already proved this model works in Bangladesh * Building a program like this for HCI is easier & cheaper than CHRF's biomedical research * HCI research, particularly in its qualitative and empirical forms, relies more on critical thinking, observation, and systematic analysis than on advanced mathematics. * Dr. Ahmed mentioned that an introduction to academia could be turned into a non-credited course at NSU if developed well. * I saved a lot of the resources that I used to learn, could be useful here. * Harvard's CS50 uses a SSG to generate and keep track of their CS50 courses. This is relatively simple to implement. I plan on keeping all notes and resources opensource and add them to DIAL's website under the subdomain. * Our demographic is likely to be T20 applicants as CHRF attracts the same types. * The west is skeptical of the work done in the global south so achievements/extracurriculars need vetted affiliations for credibility ## Why [#why] * When I first joined DIAL, I had no onboarding support. I had to "learn how to learn" research from scratch. * CHRF has the only research internship for HS students in Bangladesh, but it's biomedical. There is no equivalent for CS let alone HCI research. * **The Blocker (2024):** I pitched this in late 2024. Dr. Nova agreed in principle but identified three critical constraints: 1. **Capacity:** DIAL lacked the infrastructure to onboard 15+ students. 2. **Legitimacy:** The ECE Chair needed convincing that high schoolers could do valid research. 3. **The Trade-off:** I was tasked with building the **Research Lab Website** first to solve the legitimacy/visibility problem before we could attempt the internship. 4. I decided to also work on DIAL's foundational software and cultural architecture to solidify the ground before building something heavy on top of it. ## Related [#related] * [LabKit](/projects/labkit) — the capacity blocker; the internship waits on this being stable * [DIAL Recruitment & Onboarding Redesign](/projects/dial-recruitment) — redesigning how DIAL recruits RAs is a prerequisite for being able to onboard high schoolers * [Workshop: Writing With Authenticity in the Age of AI](/projects/writing-workshop) — planned as an optional component of the internship curriculum # Projects Overview (/projects) These projects span three contexts — academic research infrastructure, organisational systems for remote teams, and standalone experiments. Most of them share a common thread: I needed something that didn't exist, so I built it. The structure in each entry follows **What**, **Why**, and **Additional Info**. ## Research & Lab [#research--lab] Work done at or for [DIAL (North South University)](https://dial.northsouth.edu/) and adjacent academic infrastructure. ## Organisational Systems [#organisational-systems] Systems built for [Dear Asian Youth](https://dearasianyouth.org/) and South Breeze School, primarily addressing knowledge decay, credit attribution, and scaling remote teams. ## Standalone / Exploratory [#standalone--exploratory] # LabKit (/projects/labkit) *Late 2024 – Ongoing* ## What [#what] * I was asked to do a standard website refresh for DIAL when I realized that a "website" would fail because manual maintenance exacts a high cognitive tax in academia. Instead, I built LabKit, a Research Knowledge Orchestration Platform (RKOP). * Standard tools like Wix/WordPress treat research labs like a static brochure for marketing agencies and force a rigid tree structure (page -> subpage). To add a paper, the user needs to manually update the News page, the Project page, each co-author's individual profile, the Grants page, the Research Themes section and every other page that carries any relationship to that data. A single publication with two co-authors already requires five or more separate manual edits. This high maintenance tax causes content rot and students end up invisible. This is also a core ontological mismatch between the researcher's native mental model of academia, and what structure the site-builder forces them to think in. **Figure 1:** In the legacy page tree, the same paper must be entered multiple times as disconnected manual entries - once under News, once under Projects, once under each author's profile and then more. **Figure 2:** In LabKit, the same paper is a single node with declared relationships. File it once; the graph propagates it everywhere. * The redesigned system has an underlying semantic graph that has a single source of truth architecture. Researchers don't manage "pages", they input data into a "filing cabinet" (e.g., "File a Grant"). The system acts as a semantic graph that automatically propagates that connection to update the Project timeline, the Student's portfolio, and other relevant areas. The website is simply the automated frontend for this database. Filing a grant still involves declaring its relationships in which project it funds, which team members are on it but all of that happens in one context with no page-switching. ## Why [#why] * Academic content is not marketing copy. A publication is a scholarly object with strict metadata (DOI, Venue, Citations), not a blog post. Generic CMS tools break data integrity and weren't designed for academic contexts. Academic sites are often ugly but get the job done. It doesn't have to be that way. * At the current phase, this platform delivers a website. At its core, the website is just one expression of the data. By having all of the lab's data in one centralised and structured place, I am building the foundations for future agentic infrastructure. * I looked for a research lab CMS and it didn't exist. The paid website templates that existed (found only one for a lab) were incredibly ugly and didn't have half of what I needed. The rest were custom commissioned (think MIT Media Lab site). * Getting your first publication authorship on a real published paper as an undergraduate RA is difficult. In reality, RAs work on multiple projects that produce papers, and even if they don't get authorship, they get credit mentioned in the paper. The profile page for the RA in the site gives them auto-distributed credit by design rather than having to waft through papers individually. ## Additional Info [#additional-info] * **Stack:** Built on PayloadCMS 3 (headless CMS), Next.js, and MongoDB. UI components from shadcn/ui, which sets a visual standard that matches the quality of the research being produced rather than defaulting to the academic web's typical ugliness. * **Scholarly data compliance:** The site outputs JSON-LD metadata compliant with Google Scholar's indexing spec and OpenAlex's schema. DIAL's publications surface in actual academic search infrastructure instead of just Google and are readable by the AI tools that increasingly mediate how researchers discover work. I believe this will directly impact citation count and collaboration request rates for DIAL. * **Agentic SEO:** Beyond scholarly compliance, the frontend meets current agent-readability standards like server-rendered HTML with no JavaScript rendering barrier, semantic markup, clean sitemaps, and correctly configured robots.txt for AI crawlers. AI tools (GPTBot, Claude, Perplexity) can fully index the site without any additional configuration, meaning DIAL's research surfaces in AI-mediated discovery by default. * **Future direction:** The knowledge base is designed to be queryable beyond a webpage. Planned builds include a CLI and agent integration layer so AI tools can query the lab's full research graph (publications, grants, projects, team, etc) directly, without going through a page. ## Related [#related] * [ResearchNet](/projects/researchnet) — speculative automation layer that extends the data model LabKit is building * [DIAL Recruitment & Onboarding Redesign](/projects/dial-recruitment) — parallel DIAL project; LabKit needed to exist first to give DIAL the infrastructure to take on more interns * [HCI Internship for HS Students](/projects/hci-internship) — downstream: the internship program only becomes feasible once LabKit gives DIAL the capacity to handle more people * [Fieldkit OSS](/projects/fieldkit) — parallel philosophy; LabKit is to research labs what Fieldkit is to student orgs: opinionated, reproducible infrastructure # Nonbinary NLP Research Proposal (MISGENDERED++) (/projects/misgendered-nlp) Archived after *Ovalle et al. (2024)* argued that neo-pronouns were being incorrectly spliced at the tokeniser level - fixing that essentially solved the underlying engineering problem this proposal was diagnosing. A diagnostic benchmark became a solution in search of a problem. *2024 - Archived* ## What [#what] * An architectural plan for a robustness evaluation framework designed to address gaps in the *MISGENDERED* (Hossain et al., 2023) benchmark. * The paper tried to evaluate gender bias in LLMs by benchmarking how well it handled nonbinary pronoun usage. They used masked language modeling which was like a template based approach. It would feed in a sentence and ask the model to complete it (fill in the blanks) * While *MISGENDERED* relied on explicit declaration templates ("Jamie uses xe/xem"), my proposal introduced implicit contextual usage ("Jamie is preparing xyr report") and adversarial testing (injecting noise, distractors, and code-switching) to model real-world usage. * **Produced a paper proposal with citations.** This was my first independent project at DIAL, pitched to the PI and the rest of the lab. I had qualitative analysis foundations from doing IAL Psychology, I taught myself the basics of academia from YouTube and started reading papers to do a literature review. * AI research moves VERY fast, and by the time I was done learning what a literature review is in the first place, another paper came out that already solved the problem. * *Ovalle et al. (2024)* argued that neo-pronouns were being incorrectly spliced at the tokeniser level, fixing that essentially solved the problem. ## Why [#why] * I argued that existing frameworks measured memorisation of pronouns instead of *grammatical understanding*. My proposed framework would test whether the model could use the correct pronoun even when the context is noisy or ambiguous. * I archived it because I believed that since the engineering problem was being solved, a diagnostic tool (my benchmark) would be a solution in search of a problem. * I learned a lot about ML and NLP methodology, and especially at the dataset level. I read multiple papers and I remember this one case where despite not having formal ML foundations I could have a gut feeling that something won't work. * E.g, one paper argued that to reduce gender bias in LLMs, they could delete the entire vector space that gender touches. I felt like a proposal like this could only come from having a limited understanding of gender itself. It is fluid by nature and affects many things, trying to do a blunt remove is a bad idea. ## Related [#related] * [DIAL Recruitment & Onboarding Redesign](/projects/dial-recruitment) — this was my first project at DIAL; the recruitment redesign came later as I understood the lab's structural problems better # ResearchNet (/projects/researchnet) Speculative direction. The data model is being built out through [LabKit](/projects/labkit) first - ResearchNet only makes sense if the underlying data is reliable and updated. No concrete scope yet. *2025 – Ongoing* ## What [#what] * An undecided path I'm exploring to see if there's potential for more. DIAL's website has a good setup with a highly interconnected data model that's OpenAlex compliant. * Want to automate grant compliance, reporting and dissemination and explore further uses of the lab's data and especially the relationships between data * Hinges on the assumption that the data is reliable and updated ## Why [#why] * PIs often have to do a lot of administrative work that's keeping track of what went where * I don't fully understand the workflow yet so I can't design for it now but there's potential. I want to deeply understand the problem first. * You cannot have effective Agentic AI without pristine data models. I am building the full stack (website schema) myself to ensure the AI has a reliable "world model" to operate on. ## Additional Info [#additional-info] * DIAL is the perfect dogfeeding environment for this * If this works out I want to move it under Autana Systems and start doing a shift to AI/ML and especially workflow automations. I'm designing every layer of the stack so I don't have to worry about faulty data models that agentic AI needs to work with. ## Related [#related] * [LabKit](/projects/labkit) — the data model ResearchNet depends on; nothing here is scoped until that is stable # Sirkoi (/projects/sirkoi) *2024* · Exited ## What [#what] * A startup founded by friends at Colby and Pomona, originally a Rate My Professor for Bangladesh. * I named it, designed the logo, created the user journey, did the initial UI mockups in Figma, and built the website. * All of this was done pre-AI, hand-coded. ## Why I left [#why-i-left] * Only two universities in Bangladesh (NSU and BRAC) actually let students choose their professors — the model didn't map onto how the rest of the country's universities work, and I didn't think the market was there yet. * I left to focus on other work rather than keep building around that gap. * The team later explored an AI-assisted approach to generating IAL/IB study question banks. My instinct was that generating exam-quality questions needs real pedagogical expertise — I'd have leaned toward a curated question database with AI matching content to student weaknesses rather than generating questions outright. That product-judgment instinct is one I've come back to since, in how I think about [LabKit](/projects/labkit) and [ResearchNet](/projects/researchnet). # Workshop: Writing With Authenticity in the Age of AI (/projects/writing-workshop) This workshop hasn't been run yet. The outline below is a working draft - intended to be taught as part of the [HCI Internship](/projects/hci-internship) or as a standalone session. *Planned* ## What [#what] A workshop I want to teach (maybe as part of the [HCI Internship](/projects/hci-internship)). A research space I keep circling back to. Key points I want to touch on: * What constitutes authenticity and emotional honesty in writing * Why authentic writing is necessary for academia and especially qualitative research * Concrete details grounded in human experience i.e. the shared understanding of signifiers and signifieds. * Why emotion is necessary for intellectual curiosity and vitality, so it ties into quantitative research as well * i.e. you need to care about what you're doing in order to be good at it * Example: for HCI research, you need to care about marginalised communities and participants if you want to do research that makes their lives better. You need to be emotional here. * Your brain is wired towards novelty so emotion plays a role there, that is necessary for intellectual curiosity * Practical ways to practice emotional honesty in writing * How language shapes thought processes and bilingual cognition * Thinking in Bangla and translating to English is not a bad thing. It signals layered thinking and cultural context. * It signals authenticity because the thought originates from human experience with contextual layers that then gets translated. It is not an empty statement or a word mash. Bilingual cognition directly contributes towards authenticity * Example: If you're trying to learn something most likely your source material is in English. You translate it to Bangla to explain it to yourself and in the process of translation you will have processed the information and understood it better. * You can still write authentically with broken English. * **How this affects research communication** * What role does emotional honesty play in authenticity * How to not be a stochastic parrot * Examples of how LLMs can generate something that looks real but actually lacks depth in thinking. They exist in a self-referential loop of syntax without semantics. * To be a good scholar you need authenticity, as a precursor to that you need emotional honesty, for that you need emotions in the first place ## Why [#why] * The audience is going to be Bangladeshi students proficient in both English and Bangla. The postcolonial artifacts of valuing English over Bangla is widespread across South Asia and permeates personal, professional and psychological aspects of the South Asian zeitgeist. * There is this hidden shame around translating from Bangla to English in their head before speaking that should be addressed. Amy Tan's *Mother Tongue* is a great example here. * Most people are not writers like I am nor do they care as much about language as I do. However they can sense what feels authentic and what isn't. They lack the vocabulary and skills to articulate exactly why it feels that way. * Human writing is created from lived experiences, imagination or observation. AI however only stays in textual patterns. When you read a sentence and you can't close your eyes and imagine it with solid realistic details, it's likely AI generated. * AI is getting really good at faking details though. * LLMs are autoregressive models that predict the next token (\~0.75 words) based on the words that came before it. It analyses the patterns and predicts the next word in the sequence. Signifiers and signifieds hold a relationship that humans understand from experience. LLMs, however, exist in the conceptual space of signifiers from the relationships it learns during its training with no connection to reality so it cannot produce these concrete details that a human would understand. It's saying a whole lot of nothing. # elegy for friends lost to empires beyond waters unfamiliar (/writing/elegy-for-friends-lost-to-empires-beyond-waters-unfamiliar) it was the summer i became a whore\ runny sunlight like broken yolk on concrete\ my room swells like a sore throat today\ and i think of you, how do i tell you i am alive, alive\ and i want you to live too, but i miss you i listen to daniel caesar now\ and i think of you when you're not here\ love and loss never comes in stages\ i learned how to dress, did you know? longing doesn't begin to describe it\ i know you left, soon, i'll leave too\ i wish we could meet again \ my place tomorrow?\ empty chairs at the dining table all around what once was lost cannot be found\ i write pesudotranscendentalist poetry to cope\ it doesn't roll off the tongue, but it's me\ come back home one day, would you? # handprints on the unpartitioned map of india (/writing/handprints-on-the-unpartitioned-map-of-india) your eyes are black\ deep like the caves of Lascaux\ i put my palm against yours\ fingers too, soon intertwined trace the shapes in my mind, outlines\ drawing chickens, or a duck\ like children on paper\ glue on googly eyes for added effect\ crayons melting onto it, unto us\ our handprints on cave walls\ 16,000 years ago your face is a mirror, i stare into it\ my hand against yours is\ the intrinsic act of self discovery\ *but do you believe me?\ of course you do, why wouldn't you?* when i ask you to close your eyes\ do you see the face of god within me?\ in the drawn out lines of your hand\ the creases accommodate a thousand years of history do you see it like i do?\ the hazy tendrils of an afternoon in bengal\ the calmness of a moment within leaves\ swayeth gracefully; an elephantesque grandeur your skin feels familiar\ your mind a testament, to\ the unpartitioned map of india do you feel the winter creep in?\ when mornings smell like cha\ and the evenings are wrapped in shawls\ when the dust hits sharp as nicotine one of these mornings,\ when everyone else is gone, and you are too\ i will miss this\ when the silence of death is the only ringing in my ears\ i cannot fathom\ i hope the noise starts to make sense # i exist in small spaces (/writing/i-exist-in-small-spaces) i exist in the smallest of spaces\ in the underside of a bed\ a corner of a balcony the plants are my friends\ they exist in tree-time\ grounded in reality i dont want to take up more space\ i dont know how to fight\ i just want to go home sitting on a chair feels too foreign\ too grown up, fast paced\ everyone is scary i'd rather sit on the floor\ see the world through the gaps in leaves\ watch the sky grow bigger and the clouds drift apart i unbecome unto myself i miss the warmth\ my soft unbroken lips you're a child until you're not anymore\ the lines feel hazy\ i dont know where it starts or ends the facial hair feels alien\ the heaviness kicked in\ lethargic to the core\ how is it on me in the first place i am tired of fighting\ of struggling to live\ i simply wish to exist to love in simple ways\ this softness will be the death of me # Writing (/writing) A few poems and essays written over the years. # my father, drunk, talking to me in bed (/writing/my-father-drunk-talking-to-me-in-bed) dying bastard plays piano\ the curtains blindsided\ a weak yellow haze\ sugary sweet outside brown rustle,\ balm to my soul inhale:\ the heat curled around me\ like a child in\ desperation on a hot July night\ talking was simple,\ when maa wasn't there exhale:\ I breathe through heaped lungs\ your bloodshot eyes\ slurring as dawn ignites on\ the eastern skies\ caramel lights of day-break\ still stuck to my soul\ brings ants of demons / I want nothing.\ crawls and bites inhale:\ I brace for impact\ like I have done a thousand times\ before\ the blow never came\ physically, the silent tension\ I sat and let the\ words flow of your slurred speech\ I want to break free. exhale:\ your words like\ blood\ nurture my tree to grow, too fast\ too abrupt dreams of being a professor\ from mother to son to son\ seventy years the crack is where the light enters verisimilitude the broken shards of vodka glass\ like your promises, on the floor\ they are not mine.\ To give.\ Do not accept it. I cry as I watch the great\ Surrounds\ a monster's broken dreams // mirrors to bring\ to see myself as such.\ Liar. my birdsong still plays the same\ devotion always came so easy to you # Nazrul's Immaculate Rizz - My Experience with Bangla and Internet Culture (/writing/nazruls-immaculate-rizz) My relationship with the language I was born with has changed significantly over time, and this particular metamorphosis has come at a time of self-actualisation as I grow into adulthood. I strongly believe that language shapes the way we perceive the world. There is the English I speak and breathe, then there's the broken business English that my father uses to communicate emotionally difficult things because Bangla strikes too hard on the heart. There is the Bangla I hated so strongly in fourth grade and the Bangla the people in my life speak with patient, observant love. All of these languages in their many faces coexist in my life and my perception changes with every one of them. A couple of months back, I wiped the dust off my mother's dying hard drive to upload it to Google. After shifting through random wedding photos and quite a few birthdays, I found an old video of me speaking to my parents. I was four years old at the time and talked a lot. What I did notice was that my speech was entirely in Bangla. As a child, the media I consumed (over which I had no control) consisted of UNICEF's Meena cartoons and children's poetry my parents would make me learn. Another major source of content were the stories that my aunt, a cultural anthropologist, would read to me. Two years later, I would have a music teacher who would teach me how to play harmonium and sing Rabindra Sangeet, all of which I have now forgotten. It seems very impossible now, I have to put in a lot more effort to speak entirely in Bangla, or even read it. I genuinely try, but constantly find myself unable to express my thoughts the way I want to. English has become the language that I think in and exercise so much control over. I understand its tones, its nuances and cultural references. In my mind, it feels like home. It was the language that accepted me, or understood me in an intolerant society. That allowed me to exist as more than my mother's thinly veiled "do you have a *different* type of problem?" I hated learning Bangla when I started going to school. I was mostly guilted and shamed into it for the entire time I had to learn it there. We had only one class for it and had very little exposure and practice with its written form. *তোমার নিজের মাতৃভাষা তুমি শিখবা না? Won't you learn your own mother's tongue?* On the other hand, English literature classes had books so pretty they never felt like studying for me. Bangla had stories written in verbose words I was too young to comprehend. (This is why Shakespeare should be taught in tenth grade, not seventh). When I write in English, I have verisimilitude, the "truthiness" that makes you believe that what I'm saying really is true. In English, I have flow, and more importantly, experience. I have a sense of belonging. Unfortunately, with Bangla, I have none of that. I find myself having cultural clashes with my mother, someone I have been living with for the past 18 years. I feel a sort of guilt and disconnect. I struggle to understand her completely. I write unfinished poetry in a language she was beaten into memorising. Despite that, through many conversations and introspection, I have narrowed down the reason to the fact that we as a generation simply no longer occupy the cultural or literary spaces that our parents did. Instagram and TikTok (far, far more than Facebook), are a sort of abstract neighbourhood that we hang out in when we have time. As a former 14 year old, I understand the lack of demand to learn the language because Bangla simply isn't how we communicate on the internet. Our generation has moulded English into further sub-dialects that don't belong to an area and introduced nuances that would be indiscernible to the outsider who has no concept of what "rizz¹" is. Teenagers in the 80's did not speak in fluent, well articulated sentences and they will certainly not start now. Language will continue to evolve as long as people evolve. Languages are fluid and English has changed, Bangla, at least for us, has not. This abstract space that we occupy is an oversaturated representation of reality in the middle of the AI revolution. It is not a real place yet it is the unshakeable town square of our generation. It is the perfect place for us to lose our sense of self. In our hypertargerted, curated feed, we see so many people in there from around the world (mostly from the US) and so many cultures (again, US immigrants). Our desire to fit in contradicts heavily with the notion of learning a language we speak well but cannot write properly. Our consumerist, secular, modern life leaves very little scope to break from the perceived norm. The way we have built up our lives in my particular demographic, has left no room for original thought, or a moment of calm. Relaxation is essentially consuming hyper-curated media (that sometimes create echo-hells) on our phones, and all of this is in English. Language shapes thought, and thought shapes culture. And that is the one that we internalise and inhibit. On another observation, my brother is six years younger than me, on the cusp of adolescence. His clashes with my mother remind me exactly of the miscommunication that I faced when I was younger, precisely due to this cultural disconnect. Only except I had no one who would understand me at home back then. I noticed a friend of his, Samara, shortening her name to Sam, and eventually changing it to Samantha on Instagram. A complete deviation from the original Bangla. This anglicisation of her name was deliberate, yet very much a subconscious move to fit in with a growing global world. Another attempt at avoiding social rejection. My goal is not to critique social media but to provide a window for understanding and consolation into the growing cultural disconnect that I have with my parents, and that the vast majority of my generation and peers have had. It is also important that we recognize what the world is as of now. After 200 years of British rule over Bengal, and English being the language of the ruling class, it has been ingrained into our collective consciousness that English is the superior tongue. We also lose more of our own culture as we try to fit the mould of another, and our literature reflects that. Rabindranath's writing has incredible undertones of femininity in it, despite him being a bearded man. The concept of gender being intrinsically fluid in nature is prevalent all throughout his work. It is not black and white. His characters are androgynous and the ideas are far more freeing. A break from the monotony we face with the English, precisely due to the church and the state deeming all of this as savage demonic ideas from an uncivilised people. He was from a wealthy family that had supported the humanities for generations. He learnt English well enough to translate his own work, ultimately winning himself the Nobel Prize in Literature and exposing Bengali literature to the West. By comparison, Kazi Nazrul was born to Bengal's poorest of the poor. He had no interest in English. His works were later crudely translated, which the then British government used to label him as an anti-British, anti-government rebel. English was, and still is, the lingua franca for the world we live in. Ignoring it would be a grave mistake but not learning Bangla as well provides a hole for us to fall into wherein we lose our identity. Very recently have I started to reconnect with the culture I grew up in. I started watching movies by Satyajit Ray, and listening to Nazrul Geeti (which I wish I understood as well as I understand Sylvia Plath or Arundhati Roy). Because after all this time it gives me a handle to a part of my psyche I had forgotten. Self-identity always has been something every person spends their adolescence and perhaps a fair part of their twenties trying to find and establish. I enjoy listening to this music, despite not understanding it fully. I love watching these movies. It gives me a space to occupy in the abstract neighbourhood I call home. সৃজন ছন্দে আনন্দে নাচো নটরাজ হে মহকাল প্রলয়–তাল ভোলো ভোলো।।\\ * কাজী নজরুল Oh Notoraj! Dance to the joy of creation Bury your fervour of destruction * Kazi Nazrul *** Rizz ¹ - 2020s internet lingo for one’s ability to attract another person in a romantic sense. Short for “charisma” Notoraj ² - The deity of dance. A depiction of the Hindu god Shiva as the divine cosmic dancer. February 24, 2023 # the etymology of love (/writing/the-etymology-of-love) you were a mosaic of everything you have seen\ and all the people you have ever loved i. dhaka unfolds me with its stenches\ traffic and what-\ not ii. the sun spills on the sidewalk like runny yellow yolk\ folk music i hear on dirt sometimes holds\ my attention elastic iii. beneath five floors two children make a paradise out of\ one pile of sand. five fingers one hand a slap on the tyre makes it run faster iv. i see the leaves break the skin of the water we call a\ lake, unmade of tongues you pause more often now\ writing unfinished poetry\ in a language your mother was beaten into memorising\ but how do i tell you i am, i am, i am? what abrahamic glorification of poverty is enough to\ avenge the injustices of your childhood? the earth calls unto you and bangladesh is now dead\ whatever history i carry is within your memories and\ soon they will fade too\ you will forget them if you have nobody to talk to # the summer (2021) i learned what gold means (/writing/the-summer-2021-i-learned-what-gold-means) 1. it was the month my father left\ and my mother learned how to laugh again i sat down for lunch in the same\ construed table with trinkets collected over years as she and i\ pulled the afternoon apart with a fork she always wore a gold bracelet\ dulled by humid air that hung heavy\ but you're supposed to have two it was the first jewellery my father bought\ and he said he would buy her another one\ when he had the money again the years went by, their marriage fell apart\ she has not taken it off since 2001 2. two months earlier she would show me\ her jewellery box, organised, neatly placed, picked apart she said this is all i ever owned, and she picked one up\ placed it on my hand and said\ this one's for your wife, this for your brother's this one for turham's, he's like my son too\ she said, this is all i ever owned -\ i wonder if it would be different if i had daughters instead it is a strange feeling to see your parent as a person\ vulnerable, and utterly helpless - a child unburdened i don't know what it was but i cried in the kitchen sink that day 3. i was six and it was eid\ she complained that she was out of money\ i took whatever i got that day and put it in her purse she never knew it was me while i glanced\ as i made a paradise out of lego bricks\ and spoonfuls of powdered milk\ underneath the kitchen sink romeo told me in his twenties all his poems were just angry\ i think, we forget things if we have nobody to tell them to "See the gold metal I can now wear. I gave birth to your brothers and then your father gave me these two bracelets. Then I had you. And every few years, when I have a little extra money, I buy another bracelet. I know what I'm worth. They're always twenty-four carats, all genuine." — Amy Tan