WealthAgent: Multi-Agent AI for Financial Analytics

Why I Built This

Financial analysts spend a lot of time just wrangling data into the right shape to answer a question. They switch between tools, write queries, cross-reference metrics. I wanted to build something where you could ask in plain English — about trade execution, counterparty selection, or a raw database lookup — and the tool-switching happens for you.

How It Works

The system uses LangGraph and Anthropic Claude to route queries to 3 specialized agents based on intent:

• TCA Module that computes execution quality metrics: slippage, market impact, and VWAP deviation on historical trade data
• ML Recommendation Engine that analyzes historical trade performance to suggest counterparty and algorithm execution strategies
• Text2SQL Interface that converts natural language into PostgreSQL queries using LLM function calling with schema-aware prompting

Everything streams in real-time over WebSockets with live agent status indicators, so you can see which agent is working on your query as it responds.

Results

• 12+ REST endpoints with Pydantic v2 validation at all API boundaries
• 85%+ test coverage across 64 tests covering TCA computations and multi-agent orchestration
• Covers both pre-trade and post-trade analysis in a single conversational interface