Meeting Intel How it works
The processing pipeline
One background job runs these steps in order when you upload — and the page updates live as each finishes.

Upload

Text or audio.

→
OpenAI

Transcribe

Audio → text (Whisper). Only if audio.

→

Parse

Into lines: who said what, when.

→
OpenAI

Embed

Chunks → meaning numbers.

→
OpenAI

Extract

Decisions + action items.

→

Ready

Chat + panels live.

Runs in the background. A worker queue (Solid Queue) does the slow work, so the upload returns instantly.
Updates live. The status bar moves on its own (Hotwire) — transcribing → parsing → embedding → extracting → ready — no refresh.
One ordered pipeline. If a step fails, the meeting is marked failed with the reason — nothing silently half-finishes.