Aspiring 10x developer · currently benchmarking at 1.4x · Dhaka, Bangladesh
MIRAZULISLAMNAHID
Always exploring, always learning, building things as I grow.
01 — selected
Things I built
Reads the mood of your social feed. A browser extension captures the posts you scroll past, a fine-tuned xlm-roberta-base scores each one — it handles Bengali-English code-mixed text, which is most of my timeline — and a dashboard turns months of scrolling into positivity-over-time charts, busiest-hour breakdowns and keyword clouds. One account and one JWT across the extension and the web app.
A chess board that watches itself. An overhead camera on a Raspberry Pi 5 tracks a physical game and streams live algebraic notation to a built-in web UI. The trick: it never identifies piece *type*. A 3-class per-square classifier only reports empty/white/black, gated by a cheap motion detector and cross-checked against every legal move — so the software keeps piece identity and vision does the easy half.
A full-stack digital library. Read EPUB and PDF books straight in the browser, join collaborative reading rooms, get AI recommendations, and moderate the whole thing from a Filament admin panel. React 19 on Laravel 13.
A 2D light-and-shadow simulation in C with SDL2. A glowing circle casts 500 rays in every direction; a second circle drifts up and down and blocks the ones that hit it, carving out a moving shadow you can drag around with the mouse. No renderer, no GPU — every pixel is written straight into the window surface, and the circle test compares squared distances so it never calls sqrt.
02 — the person
Hello, I'm Nahid
I'm a developer in Dhaka, Bangladesh. Most of what I know came from building the thing badly first, then figuring out why it was bad. That's still the method.
My work splits about evenly between full-stack web — Laravel, Vue, React, and lately SvelteKit — and Python. When something feels like magic, I rebuild it by hand until it doesn't. That's how I ended up writing an autograd engine, and how I ended up training a sentiment model that has to cope with people typing Bengali and English in the same sentence.
I'd rather ship something small that actually works than plan something large that doesn't.
03 — the toolbox
Currently obsessed with
04 — the notebook
Lately I wrote
Our AI lab project a Chrome extension that scores every post you scroll past with a fine-tuned XLM-RoBERTa model. Soft scraping, code-mixed Bengali, and a class imbalance that almost fooled us.
How I built a push-to-deploy pipeline for a Nuxt.js app with GitHub Actions, S3, CodeDeploy and EC2 and the two errors that cost me the most time.