RAG Chatbot for Amazon 10-K Financial Analysis

This project implements a Retrieval-Augmented Generation (RAG) chatbot designed for analyzing Amazon's 10-K financial reports. The chatbot helps users efficiently obtain accurate, detailed insights from these documents.

---

Table of Contents

• Project Overview
• Key Features
• Installation
• Usage
• Enjoy the Chatbot!

---

What is RAG?

Retrieval-Augmented Generation (RAG) is an approach that combines information retrieval with the generative capabilities of Large Language Models. It works in three simple steps:

1. Retrieve: Extract documents from a knowledge base.
2. Augment: Add relevant context to the user prompt.
3. Generate: Generate an answer using the language model.

This method is particularly effective for domain-specific queries, ensuring responses are grounded in up-to-date and relevant information.

---

Project Overview

The chatbot integrates the following key components to ensure reliable performance:

• QDrant: Serves as the vector database for document indexing, retrieval, and semantic search.
• LangChain-Community: Handles chunk splitting, embeddings, and chat model integration.
• OpenAI Embeddings and Chat model: Powers answer generation.
• Gradio: Provides an easy-to-use interface for document uploads and question-answering sessions.
• RAGAS (Retrieval-Augmented Generation Assessment Score): Evaluates chatbot responses using detailed performance metrics.

The application is containerized with Docker, ensuring consistent deployment setup.

---

Key Features

1. Document Upload and Indexing

   • Upload PDF files, which are then split and embedded using OpenAIEmbeddings.

2. Vector Store with QDrant

   • Indexed documents are stored in QDrant collections, enabling fast semantic search.

3. Gradio UI

   • User-friendly chat interface that allows you to:
     • Select or create QDrant collections.
     • Ask questions with real-time references to the source text.
     • View retrieved text chunks, ensuring transparency on the source of each answer.

4. RAGAS Evaluation

   • For details on the RAGAS evaluation methodology and results, see the README.

5. Sentence-Window Retrieval

   • Functionality: Splits documents into overlapping windows based on sentences rather than arbitrary chunks.
   • Benefits: Enhances retrieval accuracy by preserving document context at the sentence level.

6. Auto-merging Retrieval

   • Benefits: Dynamically merges relevant retrieved information, delivering coherent and context-rich answers.

7. Conversational Memory

   • Benefits: Retains the context from previous interactions to support natural and continuous conversational flows.

8. Unit Tests

   • Ensures core functionalities (document indexing, retrieval, and answer generation) work as expected.

---

Installation

Follow these steps to set up the project:

1. Clone the repository:

   git clone https://github.com/rostyslavshovak/RAG-Retrieval-Augmented-Generation.git
   cd RAG-Retrieval-Augmented-Generation
   
   

2. Install required dependencies:

   pip install --no-cache-dir -r requirements.txt

3. Create a .env file in the root directory:

   touch .env

   Configure environment variables in a .env file (refer to example.env).

   OPENAI_API_KEY=sk-xxxxxx
   RAGAS_APP_TOKEN=apt.xxxxxx
   
   MODEL_NAME=gpt-3.5-turbo
   TEMPERATURE=0.0
   MAX_TOKENS=500
   EMBEDDING_MODEL=text-embedding-ada-002
   
   HOST=localhost
   PORT=6333

---

Usage

Local Setup:

1. Launch the chatbot locally:

   python -m src.gradio_ui

   • Access the application at http://localhost:7860/.

Docker Setup:

2. Run the application with Docker to ensure consistent performance:

   docker-compose up --build

   Access via http://localhost:7860/.

3. Interact with the Chatbot Use the interface to upload and index documents, then start interacting by asking specific questions

   Note: Ensure you have created or selected a QDrant collection before querying.

   • Indexing a Document:
     In the Gradio interface, click "Index a new PDF", upload your document, and specify the name for a new or existing QDrant collection.

   • Asking Questions:
     Navigate to the "Chatbot" tab, select your desired collection, and enter your questions. Relevant document chunks will appear in the Retrieved Chunks tab.

---

Enjoy the Chatbot!