Skip to content

README.md

Latest commit

 

History

History
113 lines (74 loc) · 2.66 KB

README.md

File metadata and controls

113 lines (74 loc) · 2.66 KB
Raw

💊 Pharma Chatbot

An intelligent, context-aware chatbot built to answer questions about pharmacy-related topics using documents, vector search (FAISS), and LLMs from HuggingFace. Built with LangChain, Streamlit, and Mistral 7B.

Pharma Chatbot Screenshot

🚀 Features

  • 💬 Conversational interface with memory and retrieval
  • 🔍 FAISS vector search over custom PDF documents
  • 🧠 Mistral 7B LLM via Hugging Face Inference Endpoint
  • 📄 Custom prompts to keep answers grounded in context
  • ⚡ Built-in caching for performance

🛠️ Tech Stack

📁 Project Structure

├── data/                            # PDF documents
├── vectorstore/db_faiss/           # FAISS vector store (auto-generated)
├── create_memory_for_llm.py        # Loads PDFs and creates FAISS index
├── connect_memory_with_llm.py      # Retrieval + QA chain logic
├── medibot.py                      # Streamlit UI and main app
├── requirements.txt                # Python dependencies
└── .env                            # Environment variables (HF_TOKEN)

📦 Setup & Installation

1. Clone the Repository

git clone https://github.com/yourusername/pharma-chatbot.git
cd pharma-chatbot

2. Set up Environment Variables

Create a .env file and add your Hugging Face API key:

HF_TOKEN=your_huggingface_token_here

3. Install Dependencies

pip install -r requirements.txt

4. Load and Index Your PDFs

python create_memory_for_llm.py

5. Launch the Chatbot

streamlit run medibot.py

💡 Example Prompt

"What is the recommended dosage for Paracetamol according to the document?"

⚠️ Security Warning

This project uses allow_dangerous_deserialization=True when loading the FAISS index. Only use this with trusted files as it may execute arbitrary code during deserialization.

📌 To-Do

  • Add PDF upload support from UI
  • Enable feedback on chatbot answers
  • Store chat history in a database
  • Dockerize the app

🧠 Credits

Built using:

📜 License

MIT License. See LICENSE file for details.