Building an AI system around long-term human memory.

HeartTold interviews people over time by phone or in the browser, asks follow-up questions from what they've already said, preserves their stories and makes them searchable later.

Family stories are lost when the storyteller dies, and memoir kits ask older people to write. Most of them would rather talk, and many aren't comfortable with technology.

Co-founder. I've spent more than 1,000 hours on the product: strategy, the interview workflow, the experience for older users and the build, which I direct with Claude Code.

• Phone interviews in a natural voice, so a grandparent only has to answer an ordinary call, plus browser interviews for anyone who'd rather type or record.
• Follow-up questions written live from what the person has just said, with memory of earlier sessions.
• Search across everything recorded, with answers drawn from the person's own words.
• Old documents read into a family tree.
• A memoir in three styles: as told, lightly proofread, or written up.

• AI: Claude writes the follow-up questions and the memoir. Vapi and ElevenLabs run the phone interviews. Whisper transcribes. GPT-4o vision reads documents.
• Retrieval uses embeddings with full-text search, so answers come from what the person actually said.
• Deterministic code handles accounts, scheduling, payments and storage.

1,000+ h

of my own time on the product

Source: my own estimate, Oct 2026

674

commits from March to September 2026, 575 of them mine

Source: git history, 1 Oct 2026

284

automated back-end tests

Source: test suite, 1 Oct 2026

Next.js, TypeScript, Supabase, FastAPI, Railway, Vercel, Stripe, Vapi, ElevenLabs, Whisper, Claude, GPT-4o vision