spartans
Team Spartan | Shubham Kothiya, Prem Shah, Kushal Adhyaru, Mann Nada As enterprises deploy dozens of AI agents and integrate rapidly evolving open-source repos, production failures cascade across interconnected services
Core trick
Three CrewAI agents debate root-cause hypotheses over A2A messages, and a threshold engine auto-resolves or escalates incidents.
Description
Team Spartan | Shubham Kothiya, Prem Shah, Kushal Adhyaru, Mann Nada
As enterprises deploy dozens of AI agents and integrate rapidly evolving open-source repos, production failures cascade across interconnected services faster than humans can respond.
IncidentDNA detects anomalies and runs a multi-agent AI pipeline to investigate, validate, and resolve incidents autonomously — in under 2 minutes, no human required.
Three CrewAI agents powered by Claude Sonnet 4.5 (Snowflake Cortex) collaborate via A2A communication: Agent 1 classifies severity and maps blast radius, Agent 2 investigates root cause using runbooks and past incidents, Agent 5 adversarially debates the hypothesis. A threshold engine auto-resolves or escalates to humans.
Actions are executed via Composio — Block Kit Slack alerts with approve buttons and detailed GitHub issues with fix options, all idempotent and fully auditable in Snowflake.
Stack: Snowflake, CrewAI, Composio, FastAPI, React
README excerpt
# IncidentDNA — Autonomous Incident Intelligence
> **Hackathon:** Llama Lounge × Snowflake
> **Stack:** Snowflake · CrewAI · Composio · FastAPI · React
When a deploy goes wrong, IncidentDNA detects the anomaly, runs 3 AI agents to investigate and validate the root cause, then automatically posts a Slack alert and creates a GitHub issue — all in under 2 minutes, no human required.
---
## Quick Start
```bash
# 1. Activate the virtual environment (Python 3.11 — required)
source .venv/bin/activate
# 2. Install Python deps
uv pip install -r requirements.txt
# 3. Copy credentials (already filled in)
cp .env.example .env
# 4. Verify Snowflake connection
python test_agent.py snowflake
# 5. Run the full AI pipeline (Slack + GitHub will fire)
python test_agent.py agents
# 6. Start the backend API
python -m uvicorn api:app --reload --port 8000
# 7. Start the dashboard (new terminal)
cd dashboard && npm install && npm run dev
# → http://localhost:5173
```
> ⚠️ Always use `source .venv/bin/activate` before running any Python command.
> The system Python is 3.9 and is incompatible — `.venv` uses Python 3.11.
---
## How It Works
```
GitHub push / Slack event
↓
ingestion/trigger_listener.py ← Composio WebSocket listener
↓
agents/manager.py ← Orchestrates the pipeline
├── Ag1 Detector → Severity (P1/P2/P3) + blast radius
├── Ag2 Investigator → Root cause (runbooks + past incidents + metrics)
└── Ag5 Validator