🧠 PolicyBot – RAG Assistant

A simple Python + CLI based RAG project that answers company policy questions using semantic search and an offline open-source AI model. Perfect to showcase on GitHub, Resume, and Internship submissions.

---

✨ Features

• Ask questions about company policies
• Uses Refund, Shipping & Cancellation documents
• Answers only from documents (no guessing, no hallucination)
• Works fully offline (no API keys, no billing)
• Built-in evaluation script to test accuracy

---

🛠 Tech Stack

• Python
• LangChain
• Chroma DB
• Sentence Transformers (MiniLM)
• FLAN-T5 (Open-source LLM)

---

⚙️ Setup & Run

1. Clone the repo

git clone https://github.com/your-username/policybot-rag.git
cd policybot-rag

2. Create virtual environment

python -m venv venv
venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Add policy documents

Put your policy files inside the data/ folder.

5. Build vector database

python src/ingest.py

6. Run the assistant

python src/rag.py

Ask things like:

can i cancel my order after shipping?

---

📁 Project Structure

policybot-rag/ ├─ data/ └─ src/ ├─ ingest.py` ├─ rag.py ├─ prompts.py └─ evaluate.py

---

📊 Evaluation

Run: python src/evaluate.py

You will see accuracy scores, safe refusals, and no hallucination.

---

💡 Why this project is cool

• Shows real RAG architecture
• Shows prompt engineering and evaluation
• Shows hallucination control
• Looks clean on GitHub and resumes

---

🔮 Future Improvements

• Add reranking
• Add auto evaluation
• Add citations per answer