RohanExploit/Meta-hackathon

┌─────────────────────────────────────────────────────────────┐
│                     FastAPI Server (server/app.py)           │
│  POST /reset · POST /step · GET /state · POST /evaluate     │
│  GET /tasks · GET /health · POST /live/start · /stop · ...   │
├─────────────────────────────────────────────────────────────┤
│                     Core Environment                         │
│  environment/retail_env.py   ← MultiChannelRetailEnv         │
│  environment/models.py       ← Pydantic: Action, Obs, State │
│  environment/grader.py       ← Deterministic scoring [0,1]  │
│  environment/tasks.py        ← 5 difficulty tiers            │
├─────────────────────────────────────────────────────────────┤
│              Baseline Agent (inference.py)                    │
│  OpenAI-compatible client → observe → reason → act loop      │
└─────────────────────────────────────────────────────────────┘

RetailObservation(
    day: int,                    # Current day [0, horizon)
    cash: float,                 # Available funds
    inventory: Dict[str, int],   # Current stock by product
    recent_demand_luxury: Dict,  # Rolling 3-day average of realized luxury demand
    recent_demand_budget: Dict,  # Rolling 3-day average of realized budget demand
    recent_stockouts: Dict,      # Per-product stockout count (steps with unmet demand)
    prices_luxury: Dict,         # Current luxury prices
    prices_budget: Dict,         # Current budget prices
    disruption_active: bool,     # Is a disruption currently happening?
    disruption_severity: float,  # Severity [0.0–1.0]
    market_confidence: float,    # Decaying confidence indicator [0.0–1.0]
)

score = 0.5 × profit_score + 0.3 × fill_rate_score + 0.2 × efficiency_score

adjusted_score = mean_score - variance_penalty × variance(seed_scores)

# Clone the repository
git clone https://github.com/RohanExploit/Meta-hackathon.git
cd Meta-hackathon

# Setup
python -m venv .venv
source .venv/bin/activate          # Linux/Mac
# .venv\Scripts\activate           # Windows

pip install --upgrade pip
pip install -e .

# Start the environment server
uvicorn server.app:app --host 0.0.0.0 --port 8000

export API_BASE_URL=https://router.huggingface.co/v1
export MODEL_NAME=NousResearch/Nous-Hermes-2-Mistral-7B-DPO
export OPENAI_API_KEY=<your-hf-token>

python inference.py

docker build -t retail-env .

docker run -p 8000:8000 \
  -e API_BASE_URL="https://router.huggingface.co/v1" \
  -e MODEL_NAME="NousResearch/Nous-Hermes-2-Mistral-7B-DPO" \
  -e OPENAI_API_KEY="<token>" \
  retail-env

# 1. Reset to easy task
curl -s -X POST http://localhost:8000/reset \
  -H "Content-Type: application/json" \
  -d '{"task_name": "easy", "seed": 42}'

# 2. Take an action
curl -s -X POST http://localhost:8000/step \
  -H "Content-Type: application/json" \
  -d '{"action": {"action": "order", "product": "Product_A", "quantity": 3}}'

# 3. Evaluate
curl -s -X POST http://localhost:8000/evaluate \
  -H "Content-Type: application/json" \
  -d '{"summary": {"profit": 80, "baseline_profit": 100, "total_demand": 100, "total_sales": 75}}'

python -m pytest tests/test_compliance.py -v

python test_env.py    # Environment lifecycle
python test_api.py    # API simulation

python scripts/health_check.py

python scripts/pre_submission_check.py

# Requires running server + API credentials
python scripts/benchmark_inference.py

Meta-hackathon/
├── Dockerfile                  # Root-level Docker config (HF Spaces uses this)
├── README.md                   # This file
├── app.spaces.yaml             # HF Spaces auto-config (sdk: docker, tag: openenv)
├── openenv.yaml                # OpenEnv environment metadata
├── pyproject.toml              # Dependencies + entry points
├── uv.lock                     # Locked dependency versions
├── inference.py                # Baseline LLM agent (OpenAI-compatible)
│
├── environment/                # Core environment package
│   ├── __init__.py
│   ├── models.py               # Pydantic: RetailAction, RetailObservation, RetailState
│   ├── retail_env.py           # MultiChannelRetailEnv (reset, step, state)
│   ├── grader.py               # Deterministic scoring [0, 1]
│   └── tasks.py                # 5 task definitions (easy → expert)
│
├── pipeline/                   # NVIDIA-inspired RAG pipeline (optional layer)
│   ├── config.py               # Centralized model registry + tuning knobs
│   ├── safety.py               # Llama Guard 3 input/output guardrails
│   ├── router.py               # NIM-style Router Agent (query classification)
│   ├── retriever.py            # Two-stage: FAISS dense retrieval + LLM reranking
│   ├── chain.py                # LCEL orchestrator (safety→router→retrieval→gen)
│   └── ingest.py               # SemanticChunker document ingestion
│
├── server/                     # FastAPI server package
│   ├── __init__.py
│   ├── app.py                  # 12 API endpoints + LiveRunner + SSE stream
│   └── ui/
│       ├── index.html          # Terminal-style web UI (/)
│       ├── dashboard.html      # Modern visual dashboard (/dashboard) — Chart.js charts, live runner
│       └── chart.umd.min.js    # Bundled Chart.js (served at /static/chart.js)
│
├── tests/
│   └── test_compliance.py      # 6 compliance tests
│
├── scripts/
│   ├── health_check.py         # All-in-one validation
│   ├── pre_submission_check.py # Rubric-mapped checks
│   ├── benchmark_inference.py  # Reproducibility benchmark
│   └── run_tests.py            # Isolated test runner
│
├── test_env.py                 # Environment smoke test
├── test_api.py                 # API smoke test
└── .github/workflows/ci.yml   # GitHub Actions CI