Book Recommendation System
Featured
Production-grade hybrid recommender with warm/cold support, real-time similarity search, and daily retraining with hot reloads.
Production-grade hybrid recommender with warm/cold support, real-time similarity search, and daily retraining with hot reloads.
Deploying computer vision models with NVIDIA Triton Inference Server on a home GPU rig, following a deliberate experimental arc — each experiment finds a bottleneck, fixes it, and reveals the next one — backed by real benchmark numbers.
A thread-safe C++23 logger with batched, timed async flushing and pluggable sinks — a small systems project built to get concurrency, ownership, and testing right on a fully-understood surface area.
A thread pool with work-stealing and dependency-graph task scheduling in C++23, built to internalize concurrency-safety habits — ASan/UBSan/TSan, deliberate lock scoping — rather than just getting multithreaded code to pass once.
An agentic pipeline that continuously scrapes emerging AI agent frameworks and compiles them into a queryable, interlinked knowledge base — built so coding agents past their training cutoff can stay aware of what exists.
A fine-tuned EfficientNet-V2-M that classifies street-level photos into one of Quebec's 17 administrative regions, trained on a self-collected, spatially stratified Mapillary dataset.
A cross-platform command-line tool that automates my Kaggle + LLM workflow, with project-tree exports, notebook copying, and installer builds.
A very simple but technically rich multiplayer browser game: players race to collect coins on a 2D board, with live leaderboards updating in real time.