Featured Work
AI-Powered Financial News Feed
A real-time platform built with FastAPI and PostgreSQL that combines a multi-signal vector ranking engine with a LangChain-based RAG assistant, ingesting live market data via yfinance to generate dynamic, AI-driven news summaries. Redis caching cut backend overhead by 50%, and the WebSocket layer was load-tested with Locust.
View on GitHub
Canadian Property Predictor
An end-to-end machine learning pipeline built with pandas and scikit-learn to predict Canadian housing prices, reaching an R² of 0.89. Tuned Random Forest and Gradient Boosting models to minimize MAE while cutting training time by 30%, with a Streamlit interface for interactive scenario analysis.
View on GitHub
Neural Machine Translator
An English-to-French Seq2Seq Transformer built with TensorFlow, hitting a 27.4 BLEU score on the WMT14 dataset. Implements a full encoder-decoder architecture with positional encoding, multi-head attention, and teacher forcing during training.
View on GitHub
StudySync Collaborative Hub
A real-time collaborative platform with an event-driven WebSocket architecture for concurrent group chat, improving user retention by 15% via sub-second message delivery. A database-signal-driven notification engine lifted engagement by another 15%, while backend query optimizations in Django cut round trips by 20%.
View on GitHub
Minimax Chess Engine
A Python chess engine with 100% FIDE rule compliance (castling, en passant, real-time pin detection), using Alpha-Beta pruning in a NegaMax search to cut the search space by 50% while sustaining a 3-4 ply lookahead in under a second. AI computation runs on a separate process from rendering to maintain ~60 FPS.
View on GitHub