Building evidence-backed AI products and the systems that keep people in control.
Portfolio ·
LinkedIn ·
Nereid
Nereid — trusted delivery for coding agents
Coding agents can write code faster than a reviewer can safely reconstruct it. Nereid turns a GitHub issue into a customer-run delivery with an evidence packet: the plan, changes, verification, risks, and a human decision. Repositories and Codex credentials stay in the customer environment; Nereid records evidence and never merges.
It is a technical preview built in public during the OpenAI Hackathon. The current work is making the control plane durable, reviewable, and safe enough for early design partners.
 |
 |
| Nereid Customer-controlled execution, immutable evidence, and human-only approval. |
CampaignForge AI Campaign intelligence from brief to copy and creative concepts. |
 |
 |
| CreatorKit AI Practical feedback to assess and improve content before publishing. |
Grand Slam Explorer Interactive tennis analytics built with Next.js, TypeScript, and Recharts. |
More case studies, build notes, and experiments are on my portfolio.
This is a July 2026 snapshot of the four public projects above, calculated from GitHub’s repository language data. It represents code volume—not proficiency, time, or the full range of tools I use. The source snapshot is kept alongside the graphic so the numbers can be checked and refreshed.
- Start with a real workflow and make the trade-offs visible.
- Use structured outputs, tests, and explicit guardrails around AI behaviour.
- Keep data, credentials, and approval boundaries clear.
- Ship small vertical slices, learn from them, then make the next version more durable.
TypeScript · Next.js · React · Node.js · Python · FastAPI · Postgres · Drizzle · Docker · GitHub Actions · OpenAI APIs · data pipelines and evaluation tooling