AI Market Neutral: Deterministic Control

Learn how to build production multi-agent systems for regulated finance, separating AI intelligence from deterministic authority using structured debate and strict guardrails.

Overview

This talk presents a production multi-agent system designed for regulated financial environments, where AI models generate signals but are not granted decision authority.
The system (AI Market Neutral) uses four independent AI agents to debate portfolio candidates through structured rounds (proposal, critique, defense, arbitration). However, final portfolio inclusion is determined exclusively by a deterministic risk enforcement engine.
The session will focus on:

Architectural separation between intelligence and authority
Runtime guardrails applied at the AI boundary
Deterministic constraint enforcement (sector caps, turnover limits, liquidity thresholds, earnings blackout windows)
Snapshot immutability with checksum chaining
Idempotent orchestration patterns
State synchronization and drift prevention

The demo will show real production-safe code snippets for:

Data quality gating
Hard rule validation at runtime
Constraint sequencing
Snapshot creation with artifact completeness checks
Status logic in multi-agent debate

No scoring logic, ranking weights, or proprietary signal construction will be disclosed.
The objective is to present a governance-first pattern for deploying multi-agent systems in regulated industries without granting AI sovereign decision power.

Links

Tech stack