Deploying a CNN on the Nuvoton M55M1 EVB with Ethos-U55 NPU to count elevator passengers in real time — INT8 quantized, fully on-device, targeting >95% accuracy at ≥15 FPS. Monthly mentorship from Nuvoton engineers.

Building a UART-based drive communication path as a backup alongside existing CAN infrastructure. Involves cross-team coordination with the firmware team to confirm framing specs, checksums, and hardware validation. Currently pending physical serial connection testing.

Internal event platform replacing Luma for SCE club. Contributed the GET /events/:id endpoint in Go (Gin), wiring individual event retrieval from MongoDB into the frontend. Built on Kafka + Redis + Docker for production-scale concurrency.

Instagram-style event platform for SJSU clubs — students follow clubs, browse and RSVP to events, club officers post announcements, admins moderate. Full-stack with role-based auth, RSVP logic, and waitlist handling.

Paste a messy meeting transcript, get back a structured breakdown — summary, decisions, action items with owners & deadlines, blockers, and a copy-ready follow-up message. Detects missing owners and vague tasks automatically.

Designed and shipped a real-time communication health feature for the rover teleoperation dashboard. A timestamp-based state machine (GOOD / WARN / LOST / NO DATA) gives operators a single-glance system status without parsing raw metrics. Integrated into existing React architecture using hooks — no backend changes required.

Full-stack stock monitoring system integrating Finnhub API with scheduled per-symbol polling, Express REST backend, and a dynamic React dashboard. Configured Vite proxy for CORS-free integration.

Desktop app for faculty to manage student academic profiles, programming languages, comments, and reports. My contributions: built the View Student Profiles page (v0.5), implemented Edit Student Profiles with SQLite persistence (v0.7), designed the Add Comments UI (v0.8), and built the full Reports interface with whitelist/blacklist filtering and CSV export (v0.9).

Backend pipeline scraping & normalizing 500+ job postings via BeautifulSoup, stored in a Django ORM relational schema with NLP-based keyword scoring to rank the most relevant roles first.