Pluang_AI_Knowledge_Copilot

# 🟢 Pluang Knowledge Copilot  
AI Knowledge Base & Customer Support Copilot (RAG-based)

## Overview
Pluang Knowledge Copilot is an AI-powered assistant that answers user queries **strictly based on internal Pluang documents** using a **Retrieval Augmented Generation (RAG)** architecture.

The goal of this project is to demonstrate how AI systems can be:
- grounded in trusted data
- resistant to hallucinations
- production-aware (quota, cost, reliability)
- easy to reason about and extend

This project was built as part of the **Pluang Tech Intern Assignment**.

---

## Key Features

- 📚 **Document-grounded answers only**  
  The assistant answers strictly from indexed internal documents and refuses to guess when information is missing.

- 🔍 **Retrieval Augmented Generation (RAG)**  
  Combines semantic search (FAISS + embeddings) with LLM-based reasoning.

- 🧾 **Explicit source citations**  
  Every grounded answer includes clear source references.

- 🛑 **Hallucination avoidance**  
  If the answer is not present in the knowledge base, the assistant clearly states so.

- ⚙️ **Quota-aware LLM usage**  
  Automatically falls back across multiple Gemini models when quota limits are hit.

- 🧩 **Modular, clean architecture**  
  Clear separation between configuration, retrieval, prompting, and LLM logic.

---

## Architecture Overview

**High-level flow:**

1. Internal documents are loaded and embedded using a local embedding model.
2. Embeddings are stored in a FAISS vector database.
3. User queries are converted into semantic search queries.
4. Relevant document chunks are retrieved.
5. Gemini LLM generates an answer **only from retrieved context**.
6. Sources are shown only when an answer is grounded.

---

## Tech Stack

**Frontend**
- Streamlit (chat-style UI)

**Backend / AI**
- Python
- LangChain (RAG orchestration)
- FAISS (vector store)
- Gemini Flash models (generation)
- HuggingFace sentence-transformers (local embeddings)

---

## Repository Structure
```

├── app.py          # Streamlit entry point
├── core/
│ ├── config.py        # API keys & model list
│ ├── llm.py          # Gemini model fallback logic
│ ├── vectorstore.py # FAISS + embeddings
│ └── prompt.py     # Prompt template
├── data/
│ └── mock_data.json # Internal knowledge documents
├── decision_document.md
├── requirements.txt
└── README.md

```


---

## Example Queries

**Grounded queries**
- What is the minimum amount for Pluang Gold savings?
- Is there a cooling-off period for crypto withdrawals?
- What are the fees for physical gold redemption?

**Unanswerable queries (hallucination test)**
- Who is the CEO of Pluang in 2026?
- Is Pluang regulated by SEBI?
- What is Pluang’s stock price today?

---


## Screenshots

### Grounded Answer with Source Citation
![Grounded Answer](screenshots/grounded_answer_with_source_context_1.png)
<sub>Shows a grounded response with explicit source citation.</sub>

---
![Grounded Answer](screenshots/grounded_answer_with_source_context_2.png)
<sub>Shows a grounded response with explicit source citation.</sub>

---
![Grounded Answer](screenshots/grounded_answer_with_source_context_3.png)
<sub>Shows a grounded response with explicit source citation.</sub>

---

### Hallucination Avoidance (Out-of-scope Query)
![Refusal Answer](screenshots/refusal_answer.png)
<sub>Demonstrates safe refusal when information is not present.</sub>


## How to Run Locally

```bash
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\Activate
pip install -r requirements.txt
streamlit run app.py