Things I've Built
Full-stack algorithmic trading platform. MT5 bridge streams live price data from MetaTrader 5 via a Python socket server. A FastAPI engine processes signals, a Django backend handles auth and user state, and a Next.js dashboard ties it together. Includes a Streamlit backtesting app with Monte Carlo simulation and fractional order support.
Streamlit-based backtesting workbench. Hot-loads strategy classes, supports fractional orders, multi-timeframe data merging, Monte Carlo simulation with cone charts, and a run history manager. Strategies are plain Python classes — no DSL to learn.
Physical Build A fully physical nurse-assistant robot designed in Fusion 360, 3D printed, and assembled by hand. Servo-driven arms and head, animated LCD face, Raspberry Pi brain running Gemini 2.0 Flash, Whisper voice recognition, Piper TTS, and health sensors (MAX30102 heart rate, BME280 temp, GSR stress). Inspired by working in a hospital — proof that the question is never if, only when.
Streamlit Docker control panel for self-hosted servers. Role-based auth (bcrypt), admin can see all projects while regular users only see assigned folders. Git pull → Docker build → Compose up/down — all from a browser. Used daily to manage production deployments.
Production-grade Django incident tracker for Seneca. Full lifecycle: Open → Acknowledged → In Progress → Escalated → Resolved. Auto-assigns staff on status transitions, sends async email notifications to reporters + room-group staff, tracks SLA breaches, supports L2 escalation queues.
Fully functional spreadsheet built with plain HTML, CSS, and vanilla JavaScript. Cell references, cross-cell formulae (=SUM, =AVG, =MAX), multi-sheet navigation — no frameworks. Built to prove the point.
Full terminal replacement for Windows written in Python. Custom command definitions via decorators, file encryption/decryption, shell scripting, tab completion. Built before Windows Terminal was usable.
Next-generation firewall using machine learning for behavior-based threat detection. Learns normal traffic patterns, flags anomalies in real time, adapts its decision boundary as the network evolves.
No game engine, no physics library — just requestAnimationFrame and collision math. Fully playable.