threedpt
An on-device movement lab that turns webcam pose capture into traceable observations about form, asymmetry, tempo, and joint load.
Engineer at heart โข Exploring AI, robotics & the human experience
Project archive ongoing
A working record of products, research, and experiments. Some became useful systems. Others were simply a good way to find the next question.
22 entries ยท 2024 to now
An on-device movement lab that turns webcam pose capture into traceable observations about form, asymmetry, tempo, and joint load.
Cross-session retrospectives and live coaching for people who build with Claude Code, Codex, and Gemini CLI.
A multimodal sketch-to-simulation prototype that turns drawings and text into interactive physics scenes.
An on-device movement lab that turns webcam pose capture into traceable observations about form, asymmetry, tempo, and joint load.
Cross-session retrospectives and live coaching for people who build with Claude Code, Codex, and Gemini CLI.
A Rust push-to-talk companion combining streaming transcription, screen context, Claude, and spoken replies.
A generative ecosystem of evolving plants, creatures, weather, sound, and shareable worlds, grown through a nested agent loop.
A nested autonomous improvement loop with builder and supervisor agents, file-based handoffs, and explicit validation.
A TypeScript browser-game experiment reimagining the movement and atmosphere of Prince of Persia.
An agent prototype that researches concepts, plans explanations, and generates visualizations across several rendering tools.
A multimodal sketch-to-simulation prototype that turns drawings and text into interactive physics scenes.
A Windows speech-to-text app with global hotkeys, tray controls, voice-activity detection, and local Whisper inference.
A consumer-GPU world-model prototype for generating interactive 2D environments through compact latent dynamics.
A research implementation exploring conversational speech through cross-attention, hierarchical RVQ, and prosodic memory.
An autonomous research loop that finds, verifies, and deduplicates emerging Claude Code practices into a living knowledge base.
A practical exploration of parameter-efficient fine-tuning and its memory-performance tradeoffs.
An automated research workflow for finding, extracting, and synthesizing new papers from arXiv.
A multi-step lesson-planning assistant with resource search, scoring, saved plans, and traceable generation stages.
An interactive teaching tool that makes transformer attention visible step by step.
A production-minded sentiment API built around fast inference, caching, and horizontal scale.
A production retrieval system paired with an eight-metric evaluation framework.
Research into preference optimization for vision-language reasoning, including a purpose-built multimodal CoT dataset.
A small full-stack experiment that turns a food photograph into a useful ingredient list.
An end-to-end model-serving pipeline exploring caching, health checks, and autoscaling.
A real-time computer-vision pipeline built with transfer learning on ResNet50.