The project
This prototype explores what happens when AI agents are asked to act instead of merely comment. Each agent observes a market state, builds a hypothesis, proposes an action and is evaluated against risk constraints.
Technical angle
The structure is market-data ingestion, signal normalization, specialized agents, arbitration, backtesting and result visualization. The point is not to promise that AI beats the market, but to show how autonomous decisions should be tested.
What it proves
The project demonstrates agent design with guardrails: latency, hallucination, overfitting, drawdown and evaluation all become engineering constraints.