Skip to content
Specor

Ingest Meeting Notes

The Smart Ingestions feature converts meeting transcripts, discovery session notes, or any free-form product discussion into a structured graph. The process is two phases: the AI first segments your text into discrete product topics, then generates a full node hierarchy for each topic you choose to import.

Opening Smart Ingestions

Click Smart Ingestions in the left sidebar to open the ingestions list. Click New Ingestion to start a session, or continue one you’ve saved previously.

Phase 1 — Analyze

Paste your meeting transcript or notes into the text area and click Analyze. The AI reads the full text and extracts semantic topic groups — one group per product concern (a feature to build, a user need, a rule to enforce, a flow to design, a technical constraint, or a decision made).

Each group shows:

  • Title — a short action-oriented label, e.g. “WhatsApp Quote Request Flow”
  • Summary — two sentences describing what needs to exist in the product and why it matters
  • Keywords — 3–5 classification tags
  • Source excerpt — the verbatim lines from your notes that this group is based on

Long transcripts (over 5,000 words) are processed in chunks and automatically deduplicated before results appear.

Phase 2 — Generate

Review the group cards. For each topic you want to add to your product graph, click Generate. The AI:

  1. Surveys the existing graph — searches for matching Domains, Capabilities, and Features before creating anything new, to avoid duplicates.
  2. Builds a full hierarchy — Domain → Capability → Feature, plus all satellite nodes the source text supports: Need, Actor, Rule, AcceptanceCriteria, Flow with Steps, and Decision.
  3. Fills in optional fields — scope, assumptions, rationale, impact, and other contextual fields are populated from the source excerpt where available.

A preview of the generated nodes appears on the card. Click Apply to commit the nodes to your active branch.

You can skip cards for topics you are not ready to model yet, and come back to them later.

Saving and resuming

Sessions are saved automatically as you work. Your source text, analyzed groups, and per-card state (pending / generating / applied / skipped) are all preserved so you can close the page and continue later.

Tips

  • More detail = richer output. A 40-minute transcript will produce a more complete graph than a three-line summary.
  • Group selectively. You do not have to apply every group. Skip implementation debates or out-of-scope topics you do not want in the graph yet.
  • Check for duplicates. The AI checks the graph first, but for very similar node names it may still create something new. Use the diff view after applying to catch anything unexpected.
  • Re-analyze if the notes change. Paste an updated transcript and re-run Analyze. Card states for previously applied groups are preserved.
  • Combine with AI Chat. After importing, open AI Chat to ask questions or refine the newly created nodes.