MeetingSketch

Forty minutes of technical conversation produces a usable architecture package — that is the core promise. You paste a transcript, MeetingSketch parses it for current-state components, proposed changes, implied-but-unnamed pieces, decisions made, risks flagged, and owners assigned. The vendor describes five interactive outputs per meeting: an architecture diagram (current and to-be), a decisions log, a risk register, an executive brief, and an AI Coach score. The processing is offline from your calls — no bot attends the meeting, and the vendor states your data does not train the AI model.

The differentiating feature is the AI Coach, which the vendor positions explicitly as a judgment layer rather than a summary layer. It scores the meeting against its actual objective — did this call qualify the deal, unblock the project, or move the decision? — and shows the transcript lines driving each finding. A score without evidence is just a grade; the Coach surfaces the quote that earned the flag, so you can push back on it or act on it.

The tool earns its keep in the meetings that generate architecture decisions with mixed audiences: sales discovery calls where qualification signals are buried in technical context, architecture reviews where the diagram exists only in one engineer’s head, and project kickoffs where risk owners are named once and never written down. Where it breaks: there is no API, so nothing flows downstream automatically. Teams that need diagram output inside Confluence, Jira, or a diagramming tool copy it by hand. At the free tier, processing is capped at 25 credits, which limits volume testing before a paid commitment is required.

Supported transcript sources include Otter, Fathom, Granola, Zoom, and Teams exports across eight file formats. There is no self-hosted deployment path, which means all transcript data transits the vendor’s infrastructure — a flag for teams operating under strict data residency requirements.