PROJECT.md

PROJECT.md — OpenEnv Environment Project

🎯 Project Overview

Environment Name: [ENV_NAME] Domain: [DOMAIN] (e.g., Software Engineering / Finance / Healthcare / Legal) Task Summary: [ONE_SENTENCE_DESCRIPTION_OF_REAL_WORLD_TASK]

⚠️ This must be a real-world task — not a game or toy environment.

---

🔗 Official OpenEnv References (Always Follow)

#	Tutorial
1	01-environments.md
2	02-deployment.md
3	03-scaling.md
4	04-training.md

---

🚀 High-Level Pipeline

[REAL_WORLD_TASK / DATA_SOURCE]
          │
          ▼
OpenEnv Environment  (server/environment.py)
          │   FastAPI (server/app.py)  ←→  Docker
          ▼
HTTPEnvClient (client.py)
          │   reset() / step() / state()
          ▼
GRPO Training (TRL + vLLM)
          │
          ▼
Fine-tuned LLM → pushed to Hugging Face Hub

---

📦 Tech Stack

Layer	Technology
Environment server	FastAPI + Uvicorn
Containerisation	Docker
Deployment	Hugging Face Spaces
Training framework	TRL (GRPOTrainer)
Model backend	vLLM (colocate mode)
Base model	[BASE_MODEL] (e.g., Qwen/Qwen3-1.7B)
Package manager	uv

---

📁 Repository Layout

[ENV_NAME]/
├── server/
│   ├── app.py            ← FastAPI entry point
│   ├── environment.py    ← Core environment logic
│   └── Dockerfile
├── models.py             ← Typed Action / Observation / State
├── client.py             ← HTTPEnvClient subclass
├── openenv.yaml          ← Manifest (required)
└── pyproject.toml

---

✅ Definition of Done

• openenv init scaffold created
• models.py — typed Action, Observation, State defined
• environment.py — reset(), step(), state implemented
• server/app.py — uses create_fastapi_app(env)
• curl /health → {"status": "healthy"}
• Docker image builds and runs locally
• Pushed to HF Spaces via openenv push
• GRPO training runs end-to-end
• Fine-tuned model pushed to HF Hub
• Evaluation metrics recorded