FinAgent: Intelligent Financial Question Answering

FinAgent is an intelligent assistant designed to answer financial questions by drawing information from various sources, including:

• A local SQL database (for bank account information, transactions).
• The Yahoo Finance API (for stock market data).
• A collection of PDF documents (using Retrieval Augmented Generation - RAG).

The system features a Gradio-based web interface and supports multilingual queries (English, German, Luxembourgish), translating them for internal processing and providing answers in the original language.

Key Features

• Multi-Source Question Answering: Dynamically routes questions to the most appropriate data source (SQL, API, or PDF RAG).
• Retrieval Augmented Generation (RAG): Extracts information from PDF documents to answer complex queries.
• SQL Database Interaction: Queries a local SQLite database for structured financial data.
• Live Stock Data: Fetches stock market information via the Yahoo Finance API.
• Multilingual Support: Handles questions in English, German, and Luxembourgish.
• LLM-Powered: Utilizes Large Language Models (via Ollama) for translation, classification, query generation, and natural language response generation.
• Interactive UI: Provides a user-friendly web interface built with Gradio.
• Haystack Pipelines: Built using the Haystack framework for robust NLP pipelines.

Setup & Prerequisites

1. Clone the Repository (if applicable):

   git clone <your-repository-url>
   cd FinAgent

2. Python Environment: Ensure you have Python 3.8+ installed. It's recommended to use a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies: Install the required Python packages from requirements.txt:

   pip install -r requirements.txt

4. Ollama & LLMs:

   • Install Ollama: https://ollama.com/
   • Pull the necessary LLM models used by the project (e.g., gemma3:12b):

     ollama pull gemma3:12b

5. Qdrant Vector Store: Qdrant is used for the PDF RAG pipeline. The vector store will be created locally in a qdrant/ directory when you run the indexing script.

6. Sample Database: The project expects a SQLite database named bank_demo.db in the project root for SQL-based queries. Ensure this file is present and contains the transactions table as expected by the sqlpipe function.

7. PDF Documents for RAG:

   • Place your PDF documents in the pdf_rag/data/original_pdf/ directory. These documents will be indexed for the RAG pipeline.

8. (Optional) OpenAI API Key: If you plan to use OpenAI models for the RAG pipeline (as an alternative to Ollama), ensure your OPENAI_API_KEY environment variable is set.

Running the Project

There are two main steps to get FinAgent running:

1. Index PDF Documents (for RAG pipeline)

This step parses, chunks, and indexes the content of your PDFs into the Qdrant vector store.

• The primary script for this is pdf_rag/pipelines/index_chunks.py.

• Run:

  python pdf_rag/pipelines/index_chunks.py

  This will create/update the qdrant/ folder in your project root. This step only needs to be done once or whenever you add/update PDF documents.

2. Run the FinAgent Application (Gradio UI)

Once the prerequisites are met and PDF documents (if used) are indexed, start the main application:

python haystack_yahooapi.py

This will:

• Initialize the Haystack pipelines.
• Start a Gradio web server.
• Print the URL to access the UI in your console (e.g., http://127.0.0.1:7860 or a public Gradio link if share=True).

Open the URL in your web browser to interact with FinAgent.

Project Structure

FinAgent/
├── haystack_yahooapi.py    # Main application, UI, and core logic
├── bank_demo.db            # Sample SQLite database for SQL queries
├── pdf_rag/                # Sub-module for PDF Retrieval Augmented Generation
│   ├── README.md           # README specific to the PDF RAG component
│   ├── data/
│   │   └── original_pdf/   # Place your PDF documents here
│   ├── pipelines/
│   │   ├── index_chunks.py # Script to index PDF documents
│   │   └── generate_answer.py # Script for RAG answer generation
│   └── ...
├── logs/                   # Directory for log files (created at runtime)
├── qdrant/                 # Directory for Qdrant vector store (created by indexing)
└── ... (other potential files like requirements.txt)

Logging

The application generates detailed logs in the logs/ directory, with filenames like FinAgent_log_YYYY-MM-DD_HH-MM-SS.log. These logs include Haystack component tracing and execution times.