MEng student at the University of Toronto (Data Analytics and Machine Learning). I build agentic and generative AI systems and measure them: retrieval ablations, golden sets, LLM-judge metrics and CI gates, rather than demos alone.
Market Research Agent (live demo)
A LangGraph agent over SEC filings and stock prices with hybrid retrieval, a cross-encoder reranker, trained volatility forecasters, a judge-and-retry loop, Langfuse tracing and an MCP server. The cross-encoder rerank lifted NDCG@6 from 0.44 to 0.56, faithfulness is 0.95 on a 99-question golden set, and cost is about $0.001 per question.
A multi-agent system that turns a client brief into a work breakdown structure, RAID log, RACI matrix and weekly status report, with a human-in-the-loop revision cycle and an offline demo mode.
ASL Fingerspelling Recognition (live demo)
Real-time fingerspelling to text using MediaPipe and a Conformer-Transformer, with CI and a hosted demo.
Two-Stage Recommendation System
Retrieval plus LambdaRank re-ranking pipeline on MovieLens 100K, compared against collaborative filtering baselines.
LLM and agents: LangGraph, RAG (hybrid search and reranking), Langfuse, RAGAS, DeepEval, Pydantic, Claude and OpenAI APIs
ML: PyTorch, scikit-learn, Hugging Face, MediaPipe
Backend and MLOps: Python, SQL (PostgreSQL, pgvector), FastAPI, Docker, GitHub Actions, MLflow, DVC, pytest