Autoresearch Systems Hackathon - Antler

Home repo: https://github.com/a1gmm/Autoresearch-Systems-Hackathon-Antler

This repo is the team home for the EHS Permit-Navigator hackathon project.

Project

We are building an AI-native EHS research swarm that turns a facility or project change into a defensible regulatory applicability matrix.

The demo target:

Given a Southern California manufacturing change, our agent swarm determines applicable EHS obligations, proves each row with current source evidence, and visibly fails closed when evidence is incomplete.

Start Here

Read these in order:

1. Team Share Packet
2. Two-Person Build Split
3. Hackathon Demo Design
4. Hacker Resources
5. Tool Integration Plan
6. Product and Build Plan
7. Agent Control Loop Contract
8. Test and Eval Plan
9. Harness V Tool Catalog

Tooling Story

• OpenAI Agents SDK: code-first agent loop, structured outputs, guardrails, and traces.
• Modal: dynamic research fan-out, parallel source extraction, worker isolation, timeouts, and scale-down.
• Raindrop: local trace debugging, replay, and eval creation.
• HowToEval: failure-driven eval practice and high-signal golden cases.

Repository Hygiene

• Do not commit sponsor credit codes, private API keys, .env files, source-cache credentials, or customer data.
• Keep demo fixtures small, reproducible, and safe to run without live network access.
• Treat final determinations as human-review research support, not legal advice or autonomous filing.