Show HN: How to analyze your LLM output – A behavioural health monitor for LLMs

Category: library

Tags: llm-monitoring, behavioral-analysis, ai-safety

Score: 6.5/10 (Innovation: 6, Technical: 7, Documentation: 7, Utility: 6)

PSA-core provides a multi-classifier behavioral health monitoring engine for LLM outputs, detecting issues like sycophancy, hallucination risk, and adversarial stress via posture sequence analysis. It uses lightweight models (MiniLM, ONNX) and offers a Chrome extension for real-time monitoring, making it a novel tool for improving LLM safety and reliability.

Target audience: AI/ML engineers, backend devs, data scientists

Repository: https://splabs.io · JavaScript · 3 stars

View on Hacker News