SmartChat Assistant

A powerful AI chat application built with Streamlit + LangGraph + Ollama that supports:

• Multi-threaded conversations
• PDF-based Retrieval-Augmented Generation (RAG)
• Web search
• Stock price lookup
• Calculator tool
• Persistent memory using SQLite checkpoints

---

Features

1. Multi-Threaded Chat

• Each conversation has a unique thread_id
• Threads are stored and can be reopened or deleted
• Chat history persists using LangGraph checkpoints

2. PDF RAG (Retrieval-Augmented Generation)

• Upload a PDF per thread

• Automatically:

  • Splits into chunks
  • Creates embeddings
  • Stores in FAISS vector DB

• Queries retrieve relevant chunks for accurate answers

3. Smart Context Injection (Important)

• Even if the LLM fails to call tools, document context is:

  • Retrieved manually
  • Injected into the system prompt

• Ensures reliable answers even with small models

4. Tooling Support

The assistant can use:

• rag_tool → query uploaded PDFs
• DuckDuckGoSearchRun → web search
• calculator → math operations
• get_stock_price → stock API (Alpha Vantage)

5. Streaming Responses

• Uses LangGraph streaming
• Displays tool usage in real-time

6. Persistent Memory (SQLite)

• Uses SqliteSaver

• Stores:

  • Messages
  • Tool calls
  • State transitions

---

Architecture

Frontend (Streamlit)

Handles:

• UI rendering
• Thread management
• File uploads
• Chat display
• Streaming responses

Backend (LangGraph Agent)

Handles:

• LLM reasoning
• Tool calling
• RAG retrieval
• State persistence

Flow

User Input
   ↓
chat_node (LLM)
   ↓
[optional] Tool Call
   ↓
tools node executes
   ↓
chat_node (final response)
   ↓
Stream to UI

---

Project Structure

.
├── app.py              # Streamlit frontend
├── backend.py          # LangGraph agent + tools + RAG
├── chatbot.db          # SQLite checkpoint DB
├── .env                # Environment variables
└── README.md

---

Setup Instructions

1. Clone the Repo

git clone <your-repo-url>
cd <repo>

2. Install Dependencies

pip install -r requirements.txt

3. Install & Run Ollama

Download Ollama: https://ollama.com

Pull a model:

ollama pull qwen3:4b

⚠️ Recommended: use qwen3:4b or higher (0.6b is weak for tool calling)

4. Environment Variables

Create .env:

OLLAMA_MODEL=qwen3:0.6b

5. Run the App

streamlit run app.py

---

How RAG Works

Step 1: Upload PDF

• Stored temporarily
• Loaded using PyPDFLoader

Step 2: Chunking

• RecursiveCharacterTextSplitter
• Chunk size: 1000
• Overlap: 200

Step 3: Embeddings

• Model: nomic-embed-text

Step 4: Vector Store

• FAISS in-memory store

Step 5: Retrieval

• Top-K similarity search (k=4)

Step 6: Injection

• Retrieved chunks inserted into system prompt

---

Tools Overview

1. rag_tool

rag_tool(query: str)

• Retrieves relevant document chunks
• Returns context + metadata

2. calculator

calculator(first_num, second_num, operation)

• add, sub, mul, div

3. get_stock_price

get_stock_price(symbol)

• Uses Alpha Vantage API

4. search_tool

• DuckDuckGo web search

---

Thread & Memory System

• Each thread = unique UUID

• Stored in:

  • checkpoints table
  • writes table

• Managed via SqliteSaver

Delete Thread

• Removes:

  • UI state
  • Database checkpoints

---

Future Improvements

• Fix Delete logic