Dockerfile

# FinSight v3 API — build context is this backend/ directory.
FROM python:3.11-slim

# Keep Python lean and unbuffered so logs stream to the platform immediately.
ENV PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1 \
    PIP_NO_CACHE_DIR=1 \
    HF_HOME=/opt/hf

WORKDIR /app

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# Bake the sentence-transformers model into the image so the first request
# doesn't pay a model download, and the container works even if outbound
# access to the model hub is restricted at runtime.
RUN python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')"

COPY . .

# Render (and most PaaS) inject $PORT. Bind 0.0.0.0 so it's reachable.
# Single worker on purpose: each worker loads its own PyTorch model copy.
EXPOSE 8000
CMD ["sh", "-c", "uvicorn main:app --host 0.0.0.0 --port ${PORT:-8000} --workers 1"]