Finchat

Finchat is an innovative financial chatbot application designed to provide users with accurate and insightful financial information and advice. This project leverages advanced technologies, including Retrieval-Augmented Generation (RAG) using Ollama, an intuitive user interface built with Streamlit, and efficient vector storage and search powered by Milvus.

Data Source :

BSE (Bombay Stock Exchange) Limited

Scraped using Bharat-sm-data PyPi

Demo Video:

Finchat-Demo.mp4

Image:

You can find the Hosted version here: https://finchat.sh-algo.in

Note that this application is running from my home servers so expect some downtimes

Key Features

• Advanced Financial Insights: Finchat uses state-of-the-art RAG techniques to deliver precise and contextually relevant financial information.
• User-Friendly Interface: Developed with Streamlit, Finchat offers a seamless and interactive user experience, making complex financial data accessible to all.
• Efficient Data Handling: The Milvus vector database ensures fast and accurate retrieval of financial documents and data, enhancing the chatbot's performance and reliability.

Technology Stack

• Ollama: For Managing the LLM models and provides Python client for interface with LLm models.
• Streamlit: To create an engaging and interactive web application interface.
• Milvus: A high-performance vector database for managing and querying large-scale vector data efficiently.

---

Data Flow

1. User Query: The user inputs a financial question or query through the Streamlit interface.
2. Query Processing: The query is sent to the Ollama component, where it is analyzed and broken down.
3. Vector Search: Ollama uses the processed query to perform a vector search in the Milvus database, retrieving relevant financial documents or data.
4. Response Generation: The retrieved information is used by Ollama to generate a precise and contextually relevant response.
5. Display Response: The generated response is displayed to the user through the Streamlit interface.

---

Installation and Replication

• PIP dependencies

pip3 install -r requirements.txt

• Milvus Vector database
  • You can also follow this from their website: https://milvus.io/docs/install_standalone-docker-compose.md

docker-compose up -d
(or)
docker compose up -d   

• Data Downloading

from collect_ay_pdf import FYReport

downloader = FYReport()
downloader.download_fy_report(
    bse_code='500035',
    2014,
    2023,
    'Data/AnnualReports'
)

The above code has dependency on Bhart-sm-data pip package (It's als0 created by me; checkout git/pypi for more info) install using below command (ignore if all packages are installed using requirements.txt)

pip3 install Bharat-sm-data

The above code creates PDF files in this folder ./Data/AnnualReports/500035/ and PDF files are named with {year}.pdf in that folder.

• You can construct the metadata for just ease of tracking and using pandas DFs (Completely Optional; make adjustments accordingly in ingest_doc.py in App folder in that case)

  Example for above DF (or) CSV can be found in Data/bse_metadata.csv

• Run the ingest.py for creating the Collection in milvus and chunking of the pdf and get the embedding for the text chunk and save it to Milvus Vector Store

• Install the Ollama and full the phi3 model Instructions are here

• start the streamlit server Using

streamlit run App/app.py

---

Feel free to modify this code to suit your needs. Happy coding!