Dockerfile

# --- Stage 1: Build the frontend ---
FROM node:20-alpine AS frontend-builder
WORKDIR /app/frontend
COPY frontend/package*.json ./
RUN npm install
COPY frontend/ ./
RUN npm run build

# --- Stage 2: Build the backend & package the app ---
FROM python:3.11-slim
WORKDIR /app

# Set environment variables
ENV PYTHONDONTWRITEBYTECODE=1
ENV PYTHONUNBUFFERED=1
ENV HF_HOME=/tmp/hf_cache
ENV PORT=7860

# Create writable cache directory for Hugging Face models
RUN mkdir -p /tmp/hf_cache && chmod 777 /tmp/hf_cache

# Copy backend requirements and install dependencies
COPY backend/requirements.txt ./backend/
RUN pip install --no-cache-dir -r backend/requirements.txt

# Pre-download the Hugging Face models during Docker build time 
# so the Space starts up instantly without timeout.
RUN python -c "from transformers import pipeline; \
pipeline('text-classification', model='j-hartmann/emotion-english-distilroberta-base', top_k=None); \
pipeline('ner', model='dslim/distilbert-NER', aggregation_strategy='simple')"

# Copy built frontend assets from stage 1
COPY --from=frontend-builder /app/frontend/dist ./frontend/dist

# Copy backend source code
COPY backend/ ./backend/

# Make the backend directory writable (needed for writing sentiment logs)
RUN chmod -R 777 /app/backend

# Expose port 7860 (Hugging Face Spaces default port)
EXPOSE 7860

# Run uvicorn server, binding to the port specified by Hugging Face Spaces
CMD ["sh", "-c", "uvicorn backend.main:app --host 0.0.0.0 --port ${PORT}"]