Projects
Selected
A fast, single-binary directory explorer in Go, built around how developers actually read projects. Trees that respect .gitignore, pull requests rendered as trees, commit hotspots, a suggested reading order, token-budgeted project maps for LLMs, and an MCP server so coding agents can use it directly.
An applied AI studio I founded. It designs, trains and tests custom machine learning models for startups, research labs and small teams: every engagement has a defined metric, a baseline to beat, held-out testing and a written result, and the client keeps all the code and weights.
Regime-conditioned analogue forecasting of market risk, with self-exciting tail-risk gating. A Rust + Python rebuild of a kNN, Poisson-gated trading model, re-tested under a pre-registered out-of-sample protocol across 15 new assets and redirected from forecasting direction to forecasting risk, which is where the data shows predictability.
A stochastic market forecasting platform. It models a stock’s terminal price distribution with Monte Carlo simulation (up to 1,000 paths), Geometric Brownian Motion and a 3-state Markov chain, then adjusts its risk assumptions using LLaMA 3.3 sentiment on live headlines.
A virtual study room web app for focused, shared work sessions — task tracking, live presence, and cross-network video calling built on Express.js, vanilla JS, Supabase, and WebRTC (with Cloudflare Realtime TURN for reliable connections across networks).
Atmos is a Python + FastAPI project that simulates and optimizes air quality sensor placement across the city. The system combines real-time WAQI data, traffic-weighted zone analysis, and optimization algorithms to propose an efficient sensor network that is cheaper and more effective than the existing CAAQMS stations.