Dockerfile

# Hugging Face Spaces Dockerfile
# Free tier: 2 vCPU, 16 GB RAM, 50 GB ephemeral disk
# Must listen on port 7860, runs as UID 1000

# ── Stage 1: Build frontend ─────────────────────────────────────────
FROM node:22-slim AS frontend

WORKDIR /frontend
COPY frontend/package.json frontend/package-lock.json* ./
RUN npm ci --no-audit --no-fund 2>/dev/null || npm install --no-audit --no-fund
COPY frontend/ .
RUN npm run build

# ── Stage 2: Build Python dependencies ──────────────────────────────
FROM python:3.12-slim AS builder

WORKDIR /build

RUN apt-get update && apt-get install -y --no-install-recommends \
    gcc g++ && \
    rm -rf /var/lib/apt/lists/*

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

# ── Stage 3: Download pre-built data from HF dataset ────────────────
FROM python:3.12-slim AS data

WORKDIR /data

# Install huggingface_hub for downloading
RUN pip install --no-cache-dir huggingface_hub

# Download pre-built SQLite DB and ChromaDB from HF dataset repo
RUN python -c "\
from huggingface_hub import snapshot_download; \
snapshot_download( \
    repo_id='thearkforyou/researchradar-data', \
    repo_type='dataset', \
    local_dir='/data', \
    allow_patterns=['researchradar.db', 'chroma_db/**', 'bm25_index.pkl'], \
)"

# ── Stage 4: Runtime ────────────────────────────────────────────────
FROM python:3.12-slim

WORKDIR /app

# Copy installed packages from builder
COPY --from=builder /install /usr/local

# NLTK data (punkt tokenizer for BM25)
RUN python -m nltk.downloader -d /usr/share/nltk_data punkt punkt_tab

ENV NLTK_DATA=/usr/share/nltk_data
ENV PYTHONUNBUFFERED=1
ENV PYTHONDONTWRITEBYTECODE=1

# HF Spaces runs as UID 1000 — create matching user
RUN useradd -m -u 1000 user

# Copy application code (owned by user)
COPY --chown=user src/ ./src/
COPY --chown=user scripts/ ./scripts/

# Copy frontend build
COPY --chown=user --from=frontend /frontend/dist ./frontend/dist

# Copy pre-built data from the data stage
COPY --chown=user --from=data /data/researchradar.db ./data/researchradar.db
COPY --chown=user --from=data /data/chroma_db ./data/chroma_db
COPY --chown=user --from=data /data/bm25_index.pkl ./data/bm25_index.pkl

# Model cache — download embedding model at build time for fast startup
ENV HF_HOME=/app/.cache/huggingface
RUN mkdir -p /app/.cache/huggingface && \
    python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('BAAI/bge-base-en-v1.5')" && \
    python -c "from sentence_transformers import CrossEncoder; CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')" && \
    chown -R user:user /app/.cache

# Default to Groq (cloud LLM) — key set via HF Spaces secrets
ENV LLM_BACKEND=groq
ENV SQLITE_DB_PATH=/app/data/researchradar.db
ENV CHROMA_DB_PATH=/app/data/chroma_db
ENV BM25_INDEX_PATH=/app/data/bm25_index.pkl

USER user
ENV HOME=/home/user
ENV PATH=/home/user/.local/bin:$PATH

EXPOSE 7860

CMD ["uvicorn", "src.api.app:app", "--host", "0.0.0.0", "--port", "7860"]