Index / Work
AI systems, full-stack apps, and hardware builds — each one solving a real problem.
Production System
Production retrieval-augmented AI assistant serving University of Toronto students across 19 courses. Led a zero-downtime migration of its search layer onto Postgres and cut retrieval latency ~40%.
Anthropic Claude · FastAPI · PostgreSQL
Deep Learning Capstone
LoRA fine tuned SigLIP 2 that aligns four Pokemon art styles and names in seven languages in one embedding space. Retro sprite accuracy rose from 0.48 to 0.68.
PyTorch · SigLIP 2 · LoRA
AI Project
AI financial intelligence platform that detects life events in client transactions and drafts compliant advisory briefs with a human in the loop.
Claude API · Next.js · FastAPI
Adaptive AI learning platform that personalises educational content to each student's level and learning pace in real time.
Gemini AI · React · Node.js
Identity-based academic intelligence platform that turns grades and syllabi into actionable engineering growth metrics.
TypeScript · PostgreSQL · Docker
Internship
Full-stack engineering at a high-growth scale-up — optimised backend queries to cut response latency by 35% and shipped 5 internal tooling apps to production.
Next.js · TypeScript · Supabase
Hardware
Voice activated LEGO gate on a DE1-SoC board. Bare metal RV32 C with a WM8731 audio driver, MFCC (mel frequency cepstral coefficient) feature extraction, and CNN (convolutional neural network) inference running on the device.
Bare metal C · RISC-V · DSP
Tetris written in pure Verilog on an FPGA (field programmable gate array), with PS/2 keyboard input, VGA video output, and a five state finite state machine driving play.
Verilog · FPGA
Physical Whac-A-Mole game built with Arduino, servo motors, and computer vision for real-time mole detection and scoring.
Arduino · C++ · Computer Vision