Why System Understanding Is Hard for LLMs and How to Fix It

Learn how an agent reconstructs distributed systems from code and configs, evaluating behavior under load and failure propagation, and how to improve LLM reliability through constrained extraction.

Overview

I built an agent that reads code and deployment configs from multiple repositories to reconstruct a distributed system and reason about its behavior under different scenarios.
The demo shows how the agent extracts structure from real, inconsistent codebases and evaluates system behavior under load, including failure propagation across services.

Tech stack