Financial Advisor LLM

This project was implemented with the goal of tackling financial inequality by addressing the structural disadvantages that arise from limited financial education, overwhelming and fragmented data, and unequal access to professional financial advice. High-quality financial advisors are typically accessible only to institutions/HNW individuals, reinforcing information asymmetries between these actors and the general public. This project aimed to unify and distil disparate data sources and provide tailored financial insights in an accessible and easily digestible format, thus lowering the cognitive and informational barriers faced by non-expert users and enabling more informed engagement with financial markets.

Features

• Framework: Leverages OpenAI's API and built with a FastAPI backend and a node.js frontend, using PostgreSQL for data storage.
• Tool calling: The LLM has access to tool calls to support operation, e.g., calling APIs, retrieving portfolios, etc.
• Multi-agentic framework: Uses an orchestration agent to direct requests to agents optimised for specific roles.
• Vector store retrieval: Stores regulatory filings in a vector database (ChromaDB/pgvector) for retrieval during inference.
• API calls: Can access up-to-date news, regulatory filings, and real-time data feeds via API calls.
• Generative UI: Populates UI elements in a generative fashion as directed by the LLM, useful for displaying data.
• User auth: User authentication handled via FastAPI/SQLAlchemy/PostgreSQL.

Installation and Execution

Requires local installation of node.js and Docker Compose. To install and execute (fill in your own API keys):

git clone git@github.com:j9smith/fin-llm.git
cd fin-llm
echo "OPENAI_API_KEY={YOUR API KEY HERE}" > .env
echo "REACT_API_APP_URL=127.0.0.1:8000" > frontend_app/goose/.env
npm start

The webpage can then be accessed at http://localhost:3000.

Contributors

This project was a joint effort between:

• @alexlambert1
• @mitch-c
• @j9smith