WealthAgent: Multi-Agent AI for Financial Analytics

Production-grade multi-agent AI system using LangGraph and Claude for financial analytics and TCA.

Problem Statement

Financial professionals need real-time advisory responses across portfolio management and trade analytics workflows, but existing tools lack conversational intelligence and multi-domain coverage. The challenge: build a system that can route diverse financial queries to specialized agents and deliver actionable insights.

Technical Approach

Engineered a multi-agent AI system using LangGraph and Anthropic Claude, orchestrating 3 specialized agents with LLM-based intent classification:

TCA Module — computes execution quality metrics including slippage, market impact, and VWAP deviation on historical trade data
ML Recommendation Engine — identifies optimal counterparty and algorithm execution strategies by analyzing historical trade performance and market conditions
Text2SQL Interface — converts natural language queries into structured PostgreSQL commands using LLM function calling with schema-aware prompting

Real-time WebSocket-based token streaming with live agent status indicators and persistent chat history.

Results

Full-stack application with 12+ REST endpoints and Pydantic v2 validation at all API boundaries
85%+ test coverage across a 64-test pytest suite covering TCA computations and multi-agent orchestration
End-to-end pre-trade and post-trade decision augmentation pipeline