Lahin
Zaman

An iOS app for people who lift and track what they eat — nutrition logging, hypertrophy programming, and weight-trend analysis in one place, built around the Rutgers dining database so a campus meal logs as precisely as a packaged food.

• A custom biomechanical volume-tracking algorithm routes secondary muscle fatigue, so assisting work counts without firing false overtraining alerts.
• A Node.js/Express proxy and C++ Nitro modules drive dynamic barcode resolution, Rutgers dining integration, and bidirectional Apple HealthKit sync.
• EAS pipelines ship TestFlight builds and over-the-air updates to active beta testers, with Dynamic Island widgets built as iOS UI extensions.

A local-first voice agent for macOS. The whole pipeline — listen, transcribe, reason, speak — runs on the machine, so there is no cloud round-trip and no per-minute speech-to-text bill.

• On-device transcription on Whisper via Metal with Silero VAD, paired with streaming sentence-boundary TTS: time-to-first-word fell 79%, from 18.7s to 3.9s.
• A self-repairing code subsystem makes autonomous edits safe with hash-verified auto-rollbacks. I also patched a regex bypass that was leaking MCP tool access past the permission layer.
• A macOS Electron overlay with a 460-node audio-reactive UI, TLS token authentication, and a fail-closed bind policy.

A real-time watcher for course registration. It polls a public course-availability API and tells you the moment a seat opens, by email or text, with no authentication setup on the user's end at all.

• Tracks unlimited sections campus-wide from a single API call, at sub-200ms alert latency.
• Multi-channel delivery over the Gmail API (OAuth2), Twilio, and carrier SMS gateways, all behind one operator-managed credential.
• Live instructor ratings inline with results, resolved through a name and department matching algorithm so common-name collisions never surface the wrong professor.

Finds people worth asking for a coffee chat, ranks them by whether they'll actually say yes, and drafts the ask. It fans a set of fields and positions out into dozens of narrow searches, ordered warmest first.

• Scores everyone on two axes — fit and reachability — weighting mutual connections and shared schools heavily.
• Python standard library only: no installs, no accounts, no data leaving the laptop.
• Built deliberately without scraping — it reads your own LinkedIn export or an opt-in on-screen harvester.