requirements.txt

# ── Web framework / API ────────────────────────────────────────────────
fastapi==0.110.0            # REST backend (rag_api.py)
uvicorn==0.27.1             # ASGI server to run the FastAPI app
python-multipart==0.0.9     # required by FastAPI for multipart file uploads (/upload)

# ── UI ─────────────────────────────────────────────────────────────────
streamlit==1.33.0           # web UI (streamlit_app.py)

# ── LLM provider ───────────────────────────────────────────────────────
groq>=0.4.0                 # Groq Cloud chat-completions client (streaming)

# ── RAG orchestration ──────────────────────────────────────────────────
# Pinned to the 0.1.x line on purpose: newer LangChain relocates
# HuggingFaceEmbeddings / Chroma imports and would break this code.
langchain==0.1.13
langchain-community==0.0.29
langchain-core==0.1.34

# ── Vector store + embeddings ──────────────────────────────────────────
chromadb==0.4.24            # persistent vector database
sentence-transformers==2.2.2  # local embeddings (all-MiniLM-L6-v2)
huggingface_hub==0.20.3     # PIN: sentence-transformers 2.2.2 imports cached_download,
                            # which was removed in huggingface_hub >= 0.26 (fresh-install break)
transformers==4.38.2        # transitive of sentence-transformers; pinned for reproducibility
torch==2.2.0                # transitive of sentence-transformers; pinned for reproducibility

# ── Document parsing ───────────────────────────────────────────────────
pymupdf==1.24.1             # provides `fitz` for PDF text extraction

# ── Web search fallback ────────────────────────────────────────────────
duckduckgo-search>=0.10.0   # DuckDuckGo fallback (/search)

# ── Configuration ──────────────────────────────────────────────────────
python-dotenv==1.0.1        # load GROQ_API_KEY / LLM_* from a .env file