Scribe AI
AI that reads the handwriting your ancestors left behind — turning fragile, hard-to-read historical records into searchable, understandable family history.
Scribe AI is a MyHeritage product built by a cross-functional team — not a solo project. My contribution was on the frontend: I led the technical design, built the proof of concept that validated the approach, and drove the end-to-end implementation alongside my colleagues.
What Scribe AI does
Family history lives in documents most people can't read — faded ledgers, cursive letters, gravestones, coats of arms. Scribe AI transcribes and analyzes those handwritten and printed historical records, then surfaces the insights hidden inside them.
For MyHeritage's tens of millions of users, it turns a photo of an old record into structured, searchable text — and a starting point for discovering who their ancestors were.
I owned the frontend side.
Making a complex AI capability feel intuitive inside the core product.
Led the frontend technical design for the launch — how the AI feature would slot into React/Angular flows without friction.
Refined and optimized the AI prompts iteratively to improve the quality and reliability of generated results.
Drove adoption of AI-assisted workflows across the team — new tooling and prompting practices that lifted engineering velocity.
From proof of concept to polished launch.
Build a PoC to prove it's real
Before committing the roadmap, I built the initial proof of concept to validate that the AI approach could actually read the records well enough to ship.
Tune the prompts until results held up
Iterated on the AI prompts against messy real-world handwriting — measuring output quality and tightening until transcriptions were trustworthy across record types.
Ship a genuinely intuitive UX
Implemented the full feature with the team — integrating a complex AI capability into the core platform so it feels like a natural part of exploring your family history, not a bolted-on tool.