How it works
A plain-language tour of what this app does and how it works under the hood — nine short pages.
- 1 What you do, and what you get The product at a glance: upload a meeting, and get a transcript, decisions, action items, and a chat.
- 2 How it works: two phases The two phases — what happens when you upload, and what happens when you ask a question.
- 3 Turning words into numbers What an “embedding” is, and why the number-matching happens in our database, not at OpenAI.
- 4 How an answer is built How the chat prompt is assembled, and how citations are kept honest.
- 5 Two AI jobs: summarise vs. answer Why the app uses the AI two different ways — one reads everything, one reads only what’s relevant.
- 6 The processing pipeline The steps a meeting goes through after you upload it — including audio transcription.
- 7 The system — one database The architecture: one Postgres stores everything, including the search vectors.
- 8 Why you can trust the answers The guardrails that stop the app confidently making things up.
- 9 A design decision we made A real trade-off we thought through — the judgment behind the chat.
Start at the top, or jump in anywhere — each page links to the next.