AI & product
Multi-Agent Trading Desk
12 LLM analyst agents on a LangGraph + FastAPI backend, driving rebalancing and backtesting over a real portfolio, with a Next.js retro-desktop frontend and live broker integration.
Role — Sole engineer — the agent backend and trading engine, plus the full-stack frontend.
Highlights
- 12 LLM analyst agents on a LangGraph backend, each taking a different view of the market, whose outputs are reconciled into a single trading signal.
- Live broker integration (InvertirOnline) with an auto-rebalancer that turns those signals into executable orders, and a backtesting engine to check them against history first.
- A retro desktop-in-the-browser frontend where each capability — portfolio, live market data, quick-trade, auto-trader, backtesting — is its own draggable, resizable window.
Overview
An algorithmic trading workstation built around a multi-agent LLM pipeline: 12 analyst agents run on a LangGraph + FastAPI backend, each reasoning over market data from a different angle, and their outputs are reconciled into rebalancing decisions.
Those decisions are not hypothetical — the system wires into a real Argentine broker API (InvertirOnline), so a signal becomes an executable order, with a backtesting engine to evaluate a strategy against history before any capital moves.
The frontend reframes the whole thing as a nostalgic desktop OS. Portfolio, live market data, quick-trade, auto-trader and backtesting are each their own draggable window, built on a hand-rolled Win98-style window manager.