Project 06 / 07 · Landing and demo
Forensi
Exploring better ways to debug failures in production AI agents.
- Next.js
- TypeScript
01 · Problem
Agent infrastructure can report a run as successful while the answer or action is wrong. Engineers then read long traces, retrieved documents and tool outputs by hand to find out why.
02 · Approach
Forensi is my project for that problem: read an existing trace, inspect the evidence, and form a hypothesis for the cause. The public site describes checks for grounding, tool-output interpretation, conflicting sources, freshness and authority, and calls that succeed technically but fail semantically. It is a landing page and a demo.