💊 MedAgent: A GenAI Assistant for Drug Insights

MedAgent is an agent-powered AI assistant that answers drug-related questions using Retrieval-Augmented Generation (RAG), live clinical trial data, and web search — tailored for healthcare professionals, researchers, and medical learners.

Built for interview-readiness at IQVIA, this project showcases:

• LangChain ReAct agent orchestration
• Chroma vector store indexing
• Streamlit UI deployment
• ClinicalTrials.gov API integration
• DuckDuckGo web fallback

---

✨ Features

• 🧠 LLM-Powered QA: Uses GPT-4 (via OpenAI API) to reason and answer complex drug-related questions.
• 🔍 RAG from MedlinePlus: Retrieves indexed drug usage, side effects, and precautions.
• 🧪 Live Clinical Trials: Queries real-time data from ClinicalTrials.gov via API v2.
• 🌐 Web Fallback: DuckDuckGo integration to answer open-ended or time-sensitive questions.
• 📆 Streamlit UI: Simple, user-friendly web interface.
• 🧰 Agentic Reasoning: Built with create_react_agent using LangChain’s ReAct pattern.

---

🧱 Tech Stack

Layer	Tech
💬 LLM	OpenAI GPT-4
📚 RAG	LangChain, HuggingFaceEmbeddings, Chroma
🔎 Tools	DuckDuckGoSearch, ClinicalTrials.gov API
🌐 UI	Streamlit
🐍 Env Mgmt	pipenv
📦 Packaging	.env, pipfile, Docker-ready
📀 Storage	Local CSV + vectorstore

---

📁 File Structure

medagent/
├── chroma_db/              # Vector store (if Chroma)
├── data/
│   └── medlineplus_drugs.csv
├── tools/
│   ├── clinical_trials_api.py      # API wrapper for ClinicalTrials.gov
│   ├── clinical_trial_tool.py      # LangChain Tool wrapper
│   └── drug_retrieval.py           # Vector search tool (Chroma/FAISS)
├── utils/
│   ├── ingest_embed.py             # Index CSV data to vectorstore
│   └── scrape.py                   # Scrapes MedlinePlus drug info
├── app.py                 # Streamlit frontend
├── agent_react.py         # ReAct agent setup
├── agent_executor.py      # AgentExecutor for reliable agent loop
├── .env                   # Stores OPENAI_API_KEY
├── Pipfile / Pipfile.lock
└── README.md

---

🛠️ Setup Instructions

1. Clone the repo

   git clone https://github.com/yourname/medagent.git
   cd medagent

2. Set up environment

   pipenv install
   pipenv shell

3. Set your API key Create a .env file:

   OPENAI_API_KEY=your-openai-key
   

4. Scrape and embed drug data

   python utils/scrape.py
   python utils/ingest_embed.py

5. Run Streamlit

   streamlit run app.py

---

🧪 Example Queries

• “What are the side effects of metformin?”
• “Are there ongoing clinical trials for GLP-1 receptor agonists?”
• “Latest FDA warning about Ozempic”
• “Compare ibuprofen and acetaminophen on safety”

---

🧠 Agent Flow (ReAct Pattern)

Question → Thought → Action → Observation → Thought → ... → Final Answer

Your agent dynamically chooses tools: local vectorstore, clinical trials API, or DuckDuckGo — depending on the query type.

---

🔐 Notes

• Chroma requires SQLite ≥ 3.35, so we patched it using pysqlite3-binary for Streamlit compatibility.
• Streamlit file watcher disabled to avoid torch.classes runtime error.

---

💡 Future Ideas

• Add LangGraph-style tool-switching with memory
• Enable live filtering on clinical trial results
• Add caching + logging for auditability
• Deploy via Docker or Hugging Face Spaces

---

📄 License

MIT — free for personal and commercial use.

---

🙌 Acknowledgments

• LangChain
• MedlinePlus
• ClinicalTrials.gov