Finbot - AI-Powered Financial Document Analysis Chatbot

Finbot is a tool for analysing and extracting information from complex financial documents. By leveraging LangChain, OpenAI embeddings, and Faiss indexing, Finbot provides users with a robust solution for querying financial data. The integration of Retrieval-Augmented Generation ensures that the answers are both relevant and contextually accurate, making Finbot an invaluable asset for financial analysts, investors, and other stakeholders.

Experience Finbot in action: Try the Deployed Chatbot

Click on the DEMO file in the repository to view the UI

Key Features

• AI-Driven Chatbot: Utilises OpenAI's models to understand and generate human language.
• LangChain: Integrated for managing prompts, handling document loading, and embedding text.
• Retrieval-Augmented Generation (RAG): Implements RAG to ensure that the responses are contextually relevant and grounded in the source documents.
• Faiss Index: Uses Faiss for efficient similarity search and clustering of dense vectors.
• User-Friendly Interface: Provides a seamless user experience with real-time feedback.

Components

Flask Web Application

The backend of Finbot is built using Flask, providing the essential tools to handle HTTP requests, manage file uploads, and serve the web interface.

LangChain

A framework designed to build applications that understand and generate human language, providing tools for managing prompts and integrating with various embeddings and language models.

OpenAI Embeddings

Converts textual data into high-dimensional vectors that capture semantic meaning and context, crucial for accurately addressing user queries based on the document content.

Faiss Index

A library for efficient similarity search and clustering of dense vectors, enabling rapid and precise retrieval of relevant document chunks in response to user queries.

Workflow

1. File Upload: Users upload 10-K or 10-Q financial documents in PDF format through the web interface.
2. Document Processing: The PDF is processed to extract and split the document into manageable chunks.
3. Embedding Generation: The text chunks are converted into embeddings using OpenAI's models.
4. Faiss Indexing: The embeddings are stored in a Faiss index for efficient similarity search.
5. Query Handling: Generates an embedding for the user query and retrieves the most relevant document chunks.
6. Answer Generation: Uses the retrieved chunks to generate a contextually relevant answer.

Project Directory Structure

The project directory is organised as follows:

• finenv/: The virtual environment directory containing Python dependencies.
• static/: Directory for static files like CSS and JavaScript to style the web interface.
• templates/: Contains HTML templates used by Flask to render web pages.
• .gitignore: Specifies files and directories to be ignored by Git.
• Procfile: Configuration file used by Railway for deploying the application.
• README.Rmd: This README file in RMarkdown format.
• app.py: The main application file where the Flask app and core logic reside.
• requirements.txt: Lists all the Python packages required to run the application.
• runtime.txt: Specifies the Python runtime version for deployment.
• .env: Contains environment variables such as the OpenAI API key (not committed to version control).

Deployment

The application is deployed on Railway. Railway provides a simple way to deploy, manage, and scale web applications with minimal configuration. The necessary configuration files (Procfile, runtime.txt) and environment variables are set up in the Railway project settings.

• Deployed Chatbot

Attribution

The chatbot icon used in this project is sourced from Flaticon