Case study
Project 5/5: a non‑profit mental health AI for young people
Lead engineer on an AI chat for Sekasin‑chat, built so every young person who reaches out gets support, not one in five.
- Year
- 2025
- Role
- Lead engineer
- Client
- Illusian Founder’s Office
- Stack
- TypeScript, React, OpenAI, Langfuse, PostgreSQL, Drizzle, Terraform, GCP
Why it exists
Sekasin‑chat is Finland’s leading mental health chat service for young people. It’s volunteer‑operated and in such high demand that historically only one in five people who reached out actually got through. Project 5/5, an initiative by the Ilkka Paananen Foundation together with Sekasin, in close collaboration with OpenAI, set out to close that gap with an AI‑powered chat that is a complement to human care, never a replacement.
My role
I led the engineering team for six months across the rebuild and launch. Some of the work:
- Took ownership of a legacy codebase that was already serving real users, and rewrote it from JavaScript to TypeScript end to end.
- Moved the in‑production chatbot from AWS to Google Cloud without downtime, with the infrastructure in Terraform.
- Made the model behaviour measurable. Prompting, evals and tracing with Langfuse, so changes to the conversational core could be judged against real transcripts rather than gut feeling.
- Turned a single‑tenant application into a multi‑tenant one, so the same platform could serve more than one organisation.
- Monorepo tooling (Turborepo, pnpm), PostgreSQL with Drizzle, React on the front end.
In a low‑threshold mental‑health setting, safety and escalation aren’t features; they’re the product. Most engineering decisions were shaped by that.
Result
The chat passed 10,000 users a little over 100 days after launch. Safety, ethics and the role of AI as a complement to human support stayed front and centre throughout.