Cognitive Architecture for ML Monitoring (CAMA)

GitHub: cognitive_architecture_checker
Publications: AAMAS 2025 - arXiv - PDF

LLM-based agent framework that autonomously monitors ML models in production environments. Features novel decision procedure algorithm and structured memories for interpretable model diagnosis.

• Developed in collaboration with Lenovo AI Lab
• Featured in AAMAS 2025 conference
• Reduces manual intervention by 65% while maintaining model performance
• Implements adaptive prompt architecture reducing token usage by 40%
• Successor: KC-Agent, published at IEEE COMPSAC 2026