Copilot is a real-time editor, not a transcription tool. It listens quietly, turns the conversation into clean, structured notes while it happens, and lets you fix anything it gets wrong — all without interrupting the flow.
Raw transcript comes in on the left. Copilot classifies each chunk into a structured “block” on the right. Flags anything it's unsure about. You correct only what is wrong.
During a meeting, lecture, or workshop, Copilot captures the raw conversation in real time.
Each spoken chunk is classified into one of five note types, with a confidence score.
Notes appear automatically — nothing blocks the session. If Copilot mislabels one, edit it in a sentence or a tap.
At the end, Copilot groups everything into a review: Key points, Decisions, Action items, Open questions.
Every capture is color-coded by type and given a confidence score. Blocks below the confidence threshold are flagged so you can spot and fix them quickly.
A notable statement or insight worth keeping.
Something the group agreed on.
A task with an implied or stated owner.
An open question raised but not yet resolved.
Everything else that fits a normal note.
Capture never blocks on confirmation. Each block carries a subtle Edit control — tap it to fix anything Copilot got wrong.
Describe the fix in plain language (e.g. “this was an example, not a task”) — Copilot reclassifies and rewrites.
Or tap a quick type chip to reclassify instantly.
The block updates in place and listening simply continues.
The demo runs entirely in the browser with scripted, simulated data — no real microphone, no backend, and no live AI calls yet. The “streaming” transcript and notes are played back on a timer to show the product flow end to end: let it finish, then edit anything that looks wrong. Real speech-to-text and classification come in later phases.