data/sources/          ← raw .txt files (drug labels, trials, call notes)
      │
      ▼
data/ingest.py         ← loads, splits into chunks, embeds with all-MiniLM-L6-v2
      │
      ▼
chroma_db/             ← persisted ChromaDB collections (384-dim vectors)
  ├── drug_info/
  ├── competitor_intel/
  └── pitch_content/
      │
      ▼
mcp_server/server.py   ← MCP server over stdio — exposes 4 retrieval tools
      │   (MCP JSON-RPC)
      ▼
agent/agent.py         ← LangGraph ReAct agent (ChatOllama + MCP tools)
      │
      ▼
ui/app.py              ← Gradio chat interface (browser)

# 1. Clone and enter the project
git clone <repo-url>
cd pharma-rag-mcp

# 2. Create and activate a virtual environment
python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS / Linux:
source .venv/bin/activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Configure environment (optional — defaults work out of the box)
cp .env.example .env
# Edit .env to set OLLAMA_MODEL and OLLAMA_BASE_URL if needed

# 5. Build the vector database (only needed once)
python -m data.run_ingest

python -m agent.agent

agent:
  history_window: 3   # number of prior turns to include

python -m ui.app

python -m mcp_server.server

# Evaluate all three collections
python -m eval.evaluate

# Evaluate one collection with k=5
python -m eval.evaluate --collection drug_info --k 5

agent:
  history_window: 3   # prior turns passed to LLM for multi-turn context

pharma-rag-mcp/
├── config.yaml             # Agent configuration
├── data/
│   ├── ingest.py           # IngestionPipeline class
│   ├── run_ingest.py       # CLI: build + spot-check all collections
│   └── sources/
│       ├── drug_labels/    # 11 × drug label .txt files
│       ├── clinical_trials/# 11 × clinical trial .txt files
│       └── call_notes/     # 11 × sales call note .txt files
├── rag/
│   ├── embeddings.py       # EmbeddingModel (all-MiniLM-L6-v2)
│   └── vectorstore.py      # VectorStoreManager (ChromaDB)
├── mcp_server/
│   ├── server.py           # MCP server entrypoint (stdio)
│   └── tools.py            # 4 retrieval tool definitions
├── agent/
│   └── agent.py            # PharmaAgent + CLI loop with memory
├── ui/
│   └── app.py              # Gradio chat UI
├── eval/
│   └── evaluate.py         # Hit Rate / MRR / Context Precision
├── memory/
│   └── long_term.json      # Persisted session history (auto-created)
├── chroma_db/              # Persisted vector collections (git-ignored)
├── .env.example
├── pyproject.toml
└── requirements.txt