OpenEnv Hackathon 2026

ProcureRL

ProcureRL is an OpenEnv-compatible procurement negotiation environment where a buyer LLM learns to negotiate over price, delivery, and quality against a stable scripted seller under real policy constraints.

Why it matters

Real procurement is not just price haggling. Agents must reason with incomplete information, market signals, competitor pressure, and a hard budget ceiling.

Core API

POST /reset
POST /step
GET /state/{session_id}
GET /health

Training setup

Qwen2.5-3B-Instruct + Unsloth + TRL/GRPO with shaped reward, scenario-aware prompting, and saved training plots from real runs.

Check health GitHub repo Training notebook