Production AI products, designed and built end to end, with a trust layer that makes them safe to rely on. Five shipped, three live on the OpenAI and Claude APIs, and more on the way.
Built with
The products
Not a portfolio. Five production AI products, each shipped end to end with a trust layer.
Watches trial registries and medical literature, gated so the AI can't invent a fact.
Maps any public GitHub codebase into a tour, a start-here map, and a Q&A box.
Breaks any contract into severity-flagged sections and hunts the clauses that cost people.
Audits a codebase across 19 classes of flaw, with a defensive posture throughout.
An interactive life-in-weeks visualization, rendered live on a canvas.
B2B infrastructure with tenant isolation and canaries built into every tenant.
How every build works
A signature feature and a canary in every build.
Every product carries a signature feature that is mine and a canary to catch misuse, behind a trust layer that rejects anything the source record can't back up. That is why the clinical pipeline can't surface an ID, a contact, or data that isn't real.
Always shipping
From first prototype to production. The list only grows.
AI-invented facts, across every product.
Grounding and provenance verification gate every output. The model rates and summarizes what the source already says. It never supplies the facts.
The principle
I ship AI that refuses to make things up, and I can prove it in production.
Most AI demos impress in a sandbox and fall apart in the real world. Mine are built the other way around: public URLs, real inputs, and a trust layer that fails loudly instead of inventing a confident wrong answer.
— Joshua
Global edge
Built to run anywhere. Live worldwide.
Every product ships to a global edge network, so it loads fast and stays reachable from anywhere on earth, with server-side keys and hard cost caps on every one.
AI is changing what software can do. What makes it trustworthy hasn't changed: grounded data, verified outputs, and hard limits on every call.
Every output is checked against its source before it ships.
The grounding checker and provenance verification run as a gate on the pipeline. The latest run passes all eight checks and found zero unsupported claims.
Powered by
Live deployments
Not screenshots. Real URLs you can open right now.
Open to Forward Deployed Engineer and AI Application Engineer roles, and freelance builds.