Projects
Explorations in code, research, and creativity
Formal Grammars as Protocol Buffers
Supporting a new format normally means hand-writing a parser and renderer and keeping them in sync forever. This compiler generates the whole type system from the grammar itself, so parsing and rendering fall out for free across 10+ formats, up to the ARM64 instruction set.
REST APIs as Typed gRPC Services
Integrating a third-party REST API usually means a bespoke client that rots as the API changes. This generates a typed gRPC service straight from the OpenAPI spec and regenerates from upstream, proven on the full OpenAI and Cloudflare APIs.
gRPC Embedding & LLM Serving Pipeline
Production gRPC infrastructure for text embeddings (rpcembed) and vLLM-based LLM serving, forming the AI inference stack that powers semantic search and RAG for agents at Accretional.
OpenVINO Go Bindings
No production-grade Go bindings existed for Intel's OpenVINO Runtime, which outperforms ONNX on Intel CPUs, so I wrote them from scratch: idiomatic Go/CGO with async inference and variable state, benchmarking 40% faster than ONNX Runtime.
Browser Automation over gRPC
Selector-based automation shatters the moment a page changes. This exposes the Chrome DevTools Protocol as a gRPC service and adds a vision-guided agent that acts on what a page actually looks like, so automation survives pages never built to be scripted.
Local AI Toolkit for Apple Silicon
Apple's on-device AI frameworks each need their own glue code, and most everyday tasks do not justify a cloud round-trip. This brings vision, audio, speech, and language models together into one Unix-composable, local-first CLI toolkit.
Statue Static Site Generator
Accretional's open source static site generator, built on SvelteKit and Markdown. This portfolio is built with it.
Distributed Platform Backend
The storage, search, and hosting backend that runs Accretional's cloud platform, a protobuf data store with semantic search, and the pieces that put a container or a site online.
Prompt and Workflow Tooling for an AI-Native IDE
The file-backed prompt and workflow system, and the deployment and auth integration, inside a container-based remote IDE built for working with coding agents.
CT Denoising and Explainability using OT-CycleGAN
Adapted OT-CycleGAN for denoising ultra-low-dose CT scans with explainability features using Grad-CAM and attention visualizations
Semantifly: Local RAG for Coding Agents
A command-line tool that gives AI coding agents local retrieval-augmented generation over your own code and data.
Academic Lecture Video Summariser
LLM and RAG-based system for summarizing long-form academic videos
Whispers of the Heart - AI Therapy Assistant
Python-based conversational AI system for journaling and therapeutic assistance
GeneWeaver
A parametric hardware generator for DNA sequence alignment implemented in Scala