Work
Client work and personal projects, from a tire retail platform to a RAG assistant.
E-commerce platform for a tire retailer. Admin dashboard, supplier import pipeline, inventory management, Redis-backed task queues, and Docker deployment.
A directory for AI tools: browse by category, file them into collections, search a prompt library, and add new entries through an admin panel. Dark UI with 3D scenes and scroll-driven animation.
Live healthcare platform delivering supplements and prescription items across Canada. WooCommerce backbone with 50+ products, PHIPA-compliant data handling, EMR integration, and automated prescription workflows.
A RAG system for LLQP exam candidates. It ingests the official study PDFs, embeds them into Qdrant, and answers questions with a citation back to the document and page it used. Runs as both a web widget and a Telegram bot.
Revamp of a Canadian pharmacy website: modernized infrastructure, automated form workflows that notify the clinic the moment a patient submits, and a new subdomain for fertility services.
Site for a long-standing Toronto optical clinic. Promotional popups and local SEO, aimed at bookings and walk-in traffic.
Site-wide redesign for a premium interiors brand: standardized header and footer, a new Services page, technical SEO work, and a custom 404.
Lightweight portfolio for a Toronto ceramic artist. His Etsy listings sync into the gallery, each piece linking straight through to buy.
Open-source Telegram bot that classifies music files by genre. Librosa pulls the audio features, a TensorFlow network makes the call, and the whole thing runs inside a Telegram chat.
Tagged and analyzed 2,000+ podcast episodes by guest, topics, and user interest. Built a Telegram bot with 150+ active users/month and a React + Flask web interface.
Trained and compared five models on the MNIST handwritten digit dataset, from a plain MLP up to a CNN with batch normalization and augmentation. The final submission landed in the top 5% on Kaggle.
Local Streamlit app that takes a podcast, from a YouTube URL or an audio file, through transcription, chunking, embedding, summarization, emotion profiling, keyword extraction, topic clustering, and hybrid recommendations.
Deterministic tile-based Pac-Man with five AI controllers (Random, Greedy, Ghost Avoidance, A*, Behavior Tree), a headless evaluation engine, and a Reports UI for multi-agent comparison over 1,000+ seeded episodes.
Six real-time computer vision modules built on OpenCV: edge detection, object tracking, and video stabilization, all running on a live webcam feed under 100ms latency.