& why you will ♥︎ it.
AI & LLM systems
Agents, retrieval, memory, and model training. I build the AI layer of a product — or the framework and tooling around it — down to the internals when that's what it takes.
Full-stack products
Idea to deployed, running software. I work solo and move like a small team: design, build, deploy, operate.
Systems & infrastructure
High-performance parsers, data formats, developer tooling, and the deployment and ops to keep them running in production.
Data & document AI
I build pipelines that extract information from documents and images, clean it up, and make it searchable or ready for other software.
* I specialize in making software easy to use. I take the KISS principle seriously.
& software I've built.
An LLM training engine and open research project asking whether a model can learn in essentially a single pass by reorganizing training around shared token prefixes. In-progress work, with results and open problems published honestly. trans.github.io/agpt
A Prolog-style logic language embedded directly in HTML syntax, with its predicates wired to the live DOM and the engine compiled to WebAssembly — declarative logic as a first-class part of the page itself. Its SYMBOL expression language is open source: github.com/tabcomputing/symbol
A serialization format about half the size of JSON that parses roughly 3× faster — by using the ASCII control codes that have sat in the standard since 1963 for exactly this, and were all but forgotten. Not new syntax; forgotten syntax. Implemented in seven languages. c0data.github.io
The engine that lets a model actually do work: agent loop, tool use, memory, and orchestration. It runs Arcana and Cyclops — an agent bus and an autonomous DevOps agent — which operate my production infrastructure on their own.
A semantic-memory library — vector storage, embeddings, and similarity search behind a clean API and CLI, with pluggable providers (OpenAI, Voyage). Long-term recall for AI applications. trans.github.io/memo
The most comprehensive core-extensions library for Ruby, maintained since 2004 and known across the Ruby community. Built to be cherry-picked — by gem or copy-paste — and recently revived with a fresh release. rubyworks.github.io/facets
A one-person "online circus" of roughly a dozen sub-sites: a video arcade with original games, an AI radio-drama generator, an AI stand-up stage, and more.
A YAML 1.2 parser and emitter in C11: zero-copy and SIMD-accelerated. It passes all 308 YAML test-suite cases, and on real-world benchmarks it outruns every other compliant parser, including rapidyaml. trans.github.io/yam
That's a selection. There's also Tab Phone (a voice-messaging app now in beta), Finfry (a double-entry accounting ledger), transfs (a tag-based filesystem with a FUSE interface), the Crystal FFI bindings for usearch, a word-market game, an AI-assisted job-hunting platform, and more.
More on GitHub& Why you should hire me.
I ship.
I'd rather hand you working software than a deck about it. Most of the work above is running right now.
I go end to end.
Model internals to production ops, and most of the stack in between. I don't need to hand off the hard part.
I move fast, and alone.
A year of the work above, largely from scratch, mostly solo. Give me a problem and I'll come back with something that runs.
Let's talk about what you need built.
Don't take my word for it — talk to my AI secretary. Goldie will ask what you're trying to build, write up a brief, and send it to me. She's one of the things on this site you can just try (she's the one swimming by up top).