PsychoPunch

Video Demo

About this project

We built PsychoPunch, an autonomous chaos engineering platform that evolves traditional resilience testing into a proactive, hypothesis-driven operation. While traditional tools like Netflix's Chaos Monkey rely on random fault injection like killing containers at fixed intervals, PsychoPunch utilizes an agentic brain powered by the Gemini API to analyze codebases and predict specific failure modes. The platform operates on a continuous verification loop where the agent inspects source code, formulates technical hypotheses, and executes precise tests by programmatically manipulating Docker containers and configuring Toxiproxy instances to inject live network and resource faults. To manage the massive influx of experimental data, we implemented a persistent memory layer backed by a custom BM25 Retrieval-Augmented Generation algorithm, allowing the agent to efficiently query its own historical markdown post-mortems and apply past context to new investigations. Ultimately, PsychoPunch serves DevOps and site reliability engineers by automating the discovery of critical, hidden vulnerabilities like concurrent write data loss and synchronous bottlenecks that standard testing suites inevitably miss, paving the way for a true automated reliability engineer that patches production threats before they impact users.

Gallery