Stefano Blando.
Italiano

Projects

NeurIPS 2026 Agenthon - Verifiable AI for Quantitative Finance

Autonomous agent architectures for financial market simulation, probabilistic forecasting, and verifiable inference across four competition tracks organized by SQA, CEWIT, and Bloomberg CTO Office.

The NeurIPS 2026 Agenthon (organized by the Society of Quantitative Analysts, CEWIT, and Bloomberg CTO Office) is a global competition focused on autonomous, verifiable AI agents for quantitative finance.

I am an active contender across all four challenge tracks:

  • Coding: Automated quantitative code generation, verification, and execution.
  • Forecasting: Probabilistic time-series forecasting with calibrated confidence bounds.
  • Simulation: High-throughput limit-order book market simulation using the ABIDES runtime.
  • Explainability & Verification: Verifiable, anytime-valid decision metrics and automated data leakage controls.

The system is architected as Docker-sandboxed autonomous agents equipped with strict temporal leakage guards, reproducible execution harnesses, and formal evaluation metrics suitable for mission-critical financial applications.