Hackathon build · AMD Developer HackathonAI Apps2026
Forensic intelligence for failed companies. Twenty-four specialized AI agents across four modes — Postmortem, Pre-Mortem, Founder Mode, Counterfactual — research in parallel, debate findings, and synthesize a verdict in 22 seconds on AMD MI300X.
Overview
The brief.
Forensic intelligence for failed companies. Twenty-four specialized AI agents across four modes — Postmortem, Pre-Mortem, Founder Mode, Counterfactual — research in parallel, debate findings, and synthesize a verdict in 22 seconds on AMD MI300X.
Receipts
- What works today
- Four analysis modes, each run by six specialised agents in parallel: 24 role configurations in total.
- Agents research, challenge each other's findings, then a synthesiser writes one verdict (about 22 seconds on one AMD MI300X).
- What I owned
- agent roles, orchestration, debate and synthesis flow, UI. (solo build)
- Known limits
- No benchmark of verdict quality against expert analysis; it shows orchestration, not research accuracy.
- Production boundary
- Would need a scored evaluation set of known company failures before its verdicts could be relied on.
Architecture
Company + mode
4 analysis modes
Six agents
research in parallel
Debate
challenge each finding
Synthesise
one verdict
