Customer Recovery Brain

CRM says Green; the customer story says otherwise. Customer Recovery Brain uses HydraDB retrieval and a local LLM to explain account risk with cited evidence and drafts a recovery plan for review.

Comment

posted an update —

Update since submission

  1. Agentic follow-up search. The agent now checks its own evidence for gaps. After the first HydraDB search, it plans up to two targeted follow-up searches (for example root cause or renewal status), merges in the new sources, and then writes one cited answer. The searches are shown in the UI. In testing, this pulled in up to 3 extra documents per question that the first search had missed.
  2. Richer account memory. Six more synthetic Cedar Labs sources were added to HydraDB: the billing and renewal record, a Q3 customer survey, a draft incident review, an internal escalation thread, an August QBR, and an email from the customer's operations lead. That makes 13 connected sources in total.
  3. Readable, cited answers. Recovery briefs now display as formatted text, with headings, lists and highlighted [S#] citations. The formatting is built safely, so text quoted from sources can't inject code into the page.
  4. Demo video and docs. There's an edited demo video (https://youtu.be/x3fNLfdC_8I), and the README now covers the new agent behaviour, setup and live test results.

All four demo questions still pass live, now in about 12–15 seconds, with every claim cited and all actions staged for human approval.