Skip to content

Writing

  • Background Jobs From Scratch — a seven-part engineering deep-dive: building a production background-job system (dedicated worker, transactional outbox, dead-letter queue, idempotency, security, and a money-moving cutover) from first principles. · Read the series

  • Document-Driven Development (D³) — the flagship: documentation isn't a byproduct of the work, it's the interface your AI works through. A type-partitioned doc corpus wired straight into your agents — with an open-source Claude Code skill that primes and captures it for you. · Read it · The skill

  • Claude.md Mastery — a three-part series on treating an agent's context file as a contract, not a manual: the 30-minute cleanup that cuts a real 1,014-line CLAUDE.md to 221, the nested files that auto-load per module, and the # workflow that grows the file without ever hand-editing it. · Read the series

  • ML Research, Explained — each machine-learning project I've done, rebuilt from first principles and grounded in the actual paper and code: a quantum classifier on a variational circuit, knowledge distillation for Bangla POS tagging and network-intrusion detection, blood-cell detection (and why augmentation is a modeling decision), a multi-output CNN, and ensemble startup-success prediction. Each reads cleanly for a general audience — and reveals its full derivations, ablation tables, and hyperparameters when you switch the site's lens to professor. · Read the series

  • This portfolio is now agent-operable — humans read it, agents call it, browsers operate it: an MCP server, WebMCP browser tools (live in Chrome's origin trial), an A2A agent card, and a fit-report engine that answers "does he fit?" with citations and honest gaps. · How it works

  • Machine Learning In The Realm Of Quantum: The State-Of-The-Art, Challenges, Future Vision and Applications Of It — Undergraduate thesis surveying QML models and testing two hybrid quantum-classical classifiers on MNIST (PennyLane, TensorFlow/Keras). · GitHub

  • Teaching quantum computing in a browser tab — Six scrollytelling lessons that run a real 2-qubit statevector simulator, from "what is a qubit" to quanvolution — no math prerequisites, no server, nothing faked. · Start lesson 1

  • Know which invariances are real — A lesson learned the hard way in a blood-cell detection project: augmentation policy is a modeling decision, not a preprocessing checkbox. · The paper

  • A site that publishes its own receipts — This portfolio commits its measured bundle sizes, test counts, and budgets, and renders them with the commit they were measured at. · The colophon