A working portfolio of shipped products, client systems, and labs. Some is live. Some is closed beta. Some is a lab. The status labels are there on purpose.
Live work, client work, and active builds.
AI music workstation. Describe the sound, shape the track.
A closed-beta music tool where producers work through melody, rhythm, and arrangement in conversation instead of only through menus. The local-first router keeps inference on-device when the task allows it.
demonstratesLocal-first AI router · browser audio engine · LLM-to-audio pipeline
Private recipe platform for real family workflows.
Recipe cards are digitized per family, with Google auth, magic-link invites, Postgres, Resend email, Render hosting, and an AI transcription worker running on a local node.
demonstratesFull-stack product ownership · auth · email deliverability · OCR pipeline
E-commerce + interactive 3D Gaussian-splat viewer + Stripe.
Client-commissioned mineral specimen shop with Stripe checkout and a browser-based 3D Gaussian-splat viewer embedded in product pages.
demonstratesClient delivery · Gaussian Splats · 3D in browser · Stripe integration
Interactive 3D Gaussian-splat mineral viewer.
Gaussian-splat viewer running as a Hugging Face Space and reused for the Borussia Minerals specimen workflow.
demonstratesGaussian Splat rendering · Hugging Face deployment
Desktop golf-sim companion for launch-monitor data.
Desktop companion for golf-simulator workflows: file-based shot ingest, course rendering, player profiles, club recommendations, and round persistence.
demonstratesElectron · file-based hardware ingest · desktop app deployment
WebXR VR game experiment for Quest-class browsers.
First-person VR fly-swat game exploring WebXR, procedural audio, haptics, and instanced rendering under mobile headset constraints.
demonstratesWebXR · Three.js · Quest-class constraints · VR comfort
Built for internal use. Some published.
Feedback protocol for AI agents. Encodes failure patterns and success signals. Published on PyPI.
SQLite-FTS5 personal knowledge search for local retrieval workflows.
Prompt-compression CLI experiment for lowering LLM context cost before API calls.
Draft notes on structured caching, retrieval reuse, and where token savings can be measured honestly.
Runtime rule→hook enforcement for AI agents. Blocks policy violations at execution time.