Spending my time getting machines to understand the world a little better and occasionally solve useful problems along the way.

• ♟️ Chess-llm: Fine-tune Qwen3-4B on Stockfish-labeled chess positions with 4-bit QLoRA, optional GRPO, and a local play UI on a consumer GPU.
• 📓 OpenMargam: OpenMargam helps people find the right guide for the right path.
• 👁️‍🗨️ SwarmMon: Fleet observability for ROS robots in the field live dashboard, offline detection, and alerting.
• ☎︎ CallGist: A weekly business-risk and opportunity report for owners who do not have time to review calls.
• 🕵🏼‍♂️ ProblemFinder: Find worthy pain points from various discussions - emails daily digest.
• 🗺️ funmap: Discover places worth leaving home for.
• 🕹️ VLA-inference: This is a lightweight VLA showcase for running state‑of‑the‑art robot policies without training.
• 🤖 VLA From Scratch: Vision-Language-Action policy built and trained end-to-end from scratch on a single consumer GPU, with behavior cloning, DAgger-lite post-training, and a full evaluation harness.
• 🎨 Image Generation From Scratch: Text-to-image diffusion model built end-to-end from scratch with DDPM noise scheduling, text conditioning, and a lightweight HTTP inference API.
• 🧠 LLM From Scratch: Transformer-based language model built and trained end-to-end from scratch.
• 🚗 Autonomous Driving with RL: PPO/SAC/TD3 policy gradient agents for vehicle control in simulation.
• 👁️ Stereo Visual Odometry: Camera pose estimation from stereo imagery using classical CV techniques (research, Colorado School of Mines).
• 🏭 Warehouse Automation: Autonomous robot navigation and task planning for warehouse environments.
• 🌊 Optical Flow Speed Estimation: Vehicle speed estimation from dashcam footage using optical flow on comma.ai data.