Project 05 / 07 · Landing + demo, seeking design partners
Forensi
Root-cause analysis for production AI agents: the likely cause, the evidence behind it, and the fix.
01 · Problem
Agent infrastructure can report a run as successful while the answer or action is semantically wrong. Engineers end up reading long traces, retrieved documents and tool outputs by hand to find out why.
02 · Approach
Forensi sits on top of existing traces instead of replacing them. Five detectors check grounding, tool-output interpretation, cross-source conflicts, freshness and authority, and calls that succeed technically but fail semantically. Each run gets a labelled cause such as Context conflict, with the causal chain shown as evidence.