This page describes PetConnect as of October 1, 2026. The project is archived and nothing described here is running.
In 2025 I was one of two founding engineers at PetConnect, a small startup building an AI assistant that helps pet owners decide what care a symptom needs. I wrote the product specs and the API contract, then built the backend: a TypeScript service that ran the chat through a safety filter, caught emergency keywords, and routed questions to different models by difficulty. The other founding engineer built the mobile app, which was meant to move onto this backend later. The startup wound down at the end of 2025, before the product reached users.
What it was
The assistant was for the moment a pet owner notices something wrong and doesn’t know whether it can wait. It would ask about the symptom, suggest the most likely next step, and point to a vet or an emergency line when the answer was urgent. It was education and triage, never a diagnosis, and every answer carried a disclaimer saying so.
What I built
| Piece | What it did | When |
|---|---|---|
| EEG data pipelines | Pipelines on Palantir Foundry that matched seizure annotations to the dog EEG recordings they belong to, from a research dataset | March to April 2025 [measured 2026-10-01, the pipeline repositories’ history] |
| Product specs | Vision, product requirements, architecture and the integration plan the rest of the work was built from | July to September 2025 [measured 2026-09-30, the specs repository’s history] |
| API contract | An OpenAPI description of the endpoints the mobile app would call | Written with the specs |
| Backend | An Express and TypeScript service on Supabase, calling OpenAI models | October to December 2025, 194 commits [measured 2026-09-30, the backend repository’s history] |
| Chat pipeline | One step drafted the answer, a second checked it, added the safety wording, and sent emergencies to the escalation path | Part of the backend |
| Emergency detection | Keywords for urgent symptoms that switched the answer to where to get help now | Part of the backend |
| Model routing | Simple questions went to a smaller, cheaper model, harder ones to a larger one | Part of the backend |
| Testing dashboard | A small web app for calling the API by hand while the mobile app wasn’t connected | Part of the backend repository |
| Checks | Lint, type checks and the test suite on every change, plus security scans for leaked secrets, vulnerable packages and risky code | Part of the backend repository |
The mobile app isn’t on this list because it wasn’t mine.
Status
Archived. The backend was never connected to the mobile app, and nothing here is deployed. The repositories are private, so this page has no source link.
Limits
- No users, results or usage figures exist to report, and this page doesn’t claim any.
- It describes the work in the backend, specs and EEG pipeline repositories only. What the other founding engineer built is theirs to describe.
- It was a startup role, not a paid contract.