Projects
Zetta
A minimal blog engine built on Bun featuring an integrated AI chatbot that uses the blog articles as a corpus to power its responses. Articles are plain JavaScript files on disk — each exports a metadata object and a Markdown string — and the server renders them to HTML with no database and no build step. One external dependency, the Markdown renderer.
The design goal was a system readable in one sitting. Request handling is a plain pattern-match chain rather than a middleware stack; templating is string substitution; content sync is a git clone plus a poll, with a webhook for immediate publishes. Runs traviscolbert.net.
Status: in use, actively maintained.
Zetta-AI
The chat assistant bundled into Zetta that answers questions on this site. It demonstrates tool-based retrieval — what the author calls a poor-man's RAG — where the model picks from a small set of named tools instead of querying a vector database. Curated tools return whole articles; a search tool covers everything else.
Built on Bun with no dependencies, calling the Claude API over plain HTTP so the request and response shapes stay visible in the source. Mounts into Zetta as a single request handler.
Status: in use, merged into Zetta.
Rails-to-Laravel migration
Ongoing work replacing a production Rails application with a Laravel one. Both applications run at the same time against one Postgres database and a small shared contract, which turns a risky single cutover into an incremental feature-by-feature move. The interesting engineering is in the parity checks: schema changes and data-semantic changes made on the Laravel side must not break the Rails app still serving traffic.
Status: in progress.
Distributed Production Platform as a Browser Extension
Currently developing a distributed manufacturing platform as a Chrome browser extension that turns any Windows workstation into a production node. The goal is to build a resilient swarm of manufactoring nodes that accept work tasks from a central broker. This work will revolutionize production workflows for production plants.
Status: in progress.
Enlighten - the unbiased AI development advisor
Developed a tool that helps software engineers feel more confident by providing "intent-free" commentary on AI-assisted coding sessions. This tool helps Travis make better sense out of the every-increasing stream of code that agentic tools generate.
Status: in use, actively maintained.
TodoAgent - a helpful collection of skills and commands for agentic coding
Maintains an evolving set of prompts, skills and commands that power Travis' daily coding workflow.
Status: in use, actively maintained.
Interests and directions
Current interests include:
Making AI more accessible for small businesses. This is what Travis refers to as "right-sized" solutions
Building small, dependency-free or dependency-light systems for nimble teams