Fast builds and competition work.

Built an AI-driven layout-aware RTL optimization workflow during the NYU and CogniChip hackathon, which focused on practical RTL design, verification, and evaluation in a short team-based format.

The workflow tied together simulation, synthesis, timing analysis, and LLM-guided patch suggestions using Verilator, Yosys, and OpenSTA. What made it interesting was not just the tooling itself, but the attempt to shorten the loop between seeing a hardware issue and deciding what to change next.

Built a forecast-driven flood risk system for New York City using NVIDIA PhysicsNeMo, FNO-based weather modeling, RAPIDS cuDF for geospatial processing, NYC Open Data feeds, and a lightweight vLLM-backed interpretation layer.

The project was an attempt to turn weather and geospatial inputs into a usable city-level risk workflow quickly enough for a hackathon setting. My part centered on geospatial data integration and the system architecture that connected forecast outputs, parcel-level context, and the final risk view. It was less about polishing a product and more about proving that the modeling, data processing, and mapping pieces could work together under time pressure.

Built and demoed as a working prototype during the event.

Built a Chrome extension using Gemini to identify clothing items and surface similar products from the browsing context. The initial goal was simple: make outfit discovery feel immediate instead of turning it into a manual search problem.

What made this one useful was not polish so much as direction. It was an early pass at combining browser interaction, vision, and recommendation logic in a way that felt practical enough to keep thinking about after the event.

Built and demoed as a Chrome-extension prototype during the hackathon.

Ran PyTorch DDP benchmarks and quantization experiments on Greene HPC, comparing scaling behavior across GPU setups and writing up the tradeoffs in a report with plots, screenshots, and run-level observations.

This was course-based work rather than a public competition, but it belongs on this page for the same reason as the hackathons: it was a short, focused build that sharpened how I think about performance, scaling, and hardware-aware experimentation.