Requesting Review: AI Diagnostic Agent Architecture with Evidence Reconciliation

_Kibuku · reddit · 2026-08-10

The author is designing an AI diagnostic system for business assessments and is seeking architectural peer review. The core workflow includes Intake, Conversation, Evidence extraction, Reconciliation, Validation, Findings, Human Review, and Report.

To ensure rigor, a G0 to G8 control model is proposed, requiring specific evidence or review conditions before advancing. The author asks developers with RAG/Agent experience for feedback on the architecture's viability, where to use deterministic rules, potential failure modes, and MVP scope.

Original post →

More from coding & agent

coding & agent channel →