15 AI agents that interview, listen and write, with safeguards I measured
15 AI agents that interview, listen and write memoirs, with a measured safeguard that stops the AI from inventing details.
The Life Before Me is an AI platform I co-founded. It interviews family members by voice and turns their stories into finished memoir books.
How it works
A live voice interviewer holds the conversation. Behind it, specialist agents pull out people, places and events, track the story's arc, check that the writing sounds like the storyteller, and critique each draft. About 60 AI steps go into a single book.
It doesn't make things up
Testing on 15 interview transcripts, the unguarded setup invented details in 51% of what it extracted. With the safeguard I built, it invented none. Every fact it keeps must trace back to something the person actually said.
It keeps working under load
If one AI provider hits its limit, the live conversation keeps going. Background work is queued, retried and survives a restart, so nothing is lost mid-interview.
What this means for your agent
The same disciplines go into every client agent: it only uses facts from your records, it's tested against real examples before it goes live, and it keeps working when an AI provider has a bad day.