Show HN: Hybrid Orchestrator – Reliable AI agents for finance

1 point

4 months ago

I built AI systems for banking and insurance. They failed too often. So I created a framework for human-AI teams.

The Hybrid Orchestrator has four patterns: (1) session state that survives context window limits, (2) multi-channel communication routing, (3) activity monitoring with triggers, and (4) human escalation pathways.

These patterns come from a production voice AI system for insurance applications. The code is Python, 97 tests, Apache 2.0.

I also published a research paper on TechRxiv (IEEE) describing the architecture.

Feedback welcome. What patterns do you use for reliable AI agents?