DEV Community

Godswill Francis
Godswill Francis

Posted on

How I built an AI that designs its own team to fix production incidents

The problem

It's 2am. A P0 hits. An engineer gets paged, wakes up, spends 45 minutes reading
logs and guessing. Meanwhile the business bleeds.

The real issue: human diagnosis is sequential, biased, and exhausting.

The idea

What if instead of one tired engineer, you deployed a society of AI specialists
that challenged each other's conclusions before acting?

That question became CHIMERA — built in 23 days on Qwen Cloud.

What makes it different

Every other agent system has hardcoded roles. CHIMERA has none.

Given any incident, Qwen Max reads it and invents the right specialists on the fly.
A database incident gets a DbConnectionAnalyst and a DeploymentForensicsEngineer.
A cache incident gets completely different agents. Zero templates.

Then — and this is the part I'm most proud of — every specialist gets a paired
adversarial agent whose only job is to attack their conclusions. No confirmation
bias. No rubber-stamping.

Building on Qwen Cloud

Qwen Max handles the high-stakes decisions — topology synthesis and consensus.
Qwen Plus runs the specialist agents in parallel. text-embedding-v3 powers the
pgvector memory system that makes CHIMERA get faster with every incident.

The Qwen API's OpenAI-compatible interface meant I could build the full multi-turn
tool execution loop cleanly — agents call real tools (SQL queries, health checks,
log analysis), observe results, and reason further until they reach a conclusion.

The result

  • P0 incident resolved in 37 seconds vs 45-minute industry average
  • 72× faster on recalled incident classes (memory system kicks in)
  • 85%+ confidence after adversarial validation
  • Zero human action required when auto-detect mode is on

What I learned

Adversarial agents change the final decision in ~30% of cases. Without them,
specialists over-confirm their first hypothesis every time. The adversarial loop
isn't just a gimmick — it's the most important architectural decision in the system.

Try it

GitHub: https://github.com/willzfrank/chimera
Demo: https://youtu.be/D_LzqNMSU18

Built with Qwen Cloud · Alibaba Cloud · NestJS · pgvector · React Flow

Top comments (0)