What does Apex Replicant extract from every session, and how does refinement work?

The problem this solves

You run a session. It mostly went well. There were two moments where the protégé got it almost right: missed a nuance, used a phrase you wouldn't, over-explained something simple.

Most platforms leave you with the transcript and nothing else. You're expected to read 30 pages, identify the issues, and write a prompt patch. That's work you won't do consistently. So proteges drift; they don't get worse exactly, they just don't get better.

Session insights + our refinement loop (feedback-to-redeploy) is the refinement loop that closes the gap. The AI does the reading. You do the judgment call. The system does the redeploy.

How it works

1. Session ends. Transcript is stored, PII is redacted.
2. Claude extracts seven categories. Sentiment trend across the session. Topics covered. Action items the client left with. Next steps the protégé recommended. Objections or skepticism the client raised. KB gaps the protégé encountered. Refinement suggestions specific to your protégé's behavior.
3. Session detail modal. You see all seven categories in a single view, with the transcript alongside.
4. Weekly digest email. Monday 6:45am Central Time. Your week in aggregate: total sessions, per-protégé breakdown, top signals. Individual per-protégé digests go out at 8am.
5. Submit feedback in plain English. Describe what you'd change. "It should have asked about funding stage before recommending a raise strategy."
6. Gemini extracts structured instructions. Your plain-English feedback becomes weighted instructions that shape the protégé.
7. Auto-regen + auto-deploy (our refinement loop). If the AI extracted at least one instruction, the protégé regenerates and pushes to our voice platform in the same request. Next session uses the new version.
8. AI-surfaced refinement AI suggestions. Monthly, Claude reviews your recent sessions against your feedback history and surfaces refinement suggestions you haven't acted on yet. Apply with one click.

How to turn it on

Session insights are on by default. The Session Detail modal is in your expert dashboard under Sessions. The Feedback tab is in your protégé profile with four sub-tabs: Submit · Active · History · Suggestions.

To make the most of it:

1. Read your Monday digest weekly
2. Click into any flagged session from the digest
3. Use the Submit tab to describe any refinement in plain English
4. Check Suggestions monthly for AI-surfaced patterns

FAQ

What happens to my session transcripts? Stored in your expert account with PII redacted. Used for insight extraction and (if you opt in) for ingestion back into your KB as a source.

Can I opt out of Claude analyzing my sessions? No. It's core to the platform. If you don't want your transcripts analyzed, don't use a voice-first session platform.

How accurate is the insight extraction? Seven categories is a deliberately tight schema. Claude does well on structured extraction like this. If you spot a miscategorized signal, the feedback loop closes the gap.

Does our refinement loop (auto-regen) break my protégé? Only in the ways you told it to. The regeneration reflects your feedback instructions. If you say "never recommend a specific brand," the protégé stops recommending that brand. If that breaks something, remove the instruction from your Active Instructions tab.

How is this different from Delphi Insights? Delphi Insights surfaces trends and signals across sessions. We do that too. What we also ship is the refinement action: your feedback directly regenerates and redeploys the protégé. Delphi's refinement is a settings-panel edit. Ours is "say what you'd change; we ship it."

What about the weekly digest? Monday 6:45am Central is your expert digest. Monday 8am Central is your per-protégé digest (one per protégé, if you run several). You can turn these off, but most experts report they're the single most-read email they get from us.