Our research in safe, explainable agentic AI
Recent papers, preprints, and ongoing work from our lab — multi-agent actor–critic systems for fault detection, reasoning, and verification.
Papers and preprints
Fault detection, severity and cause analysis in network telemetry
Actors and critics powered by Large Language Models (LLMs) in a multi-agent AI system can detect faults, estimate severity, and identify likely causes in network telemetry.
In collaboration with the University of AlbertaMultimodal task orchestration in agentic AI systems
AI agents collaborate across modalities to detect anomalies, prioritize events, and escalate critical cases while respecting real-time resource constraints.
In collaboration with the University of Alberta Research brief and explainer video coming soonOur Projects
Applied research projects from our lab.
Multimodal Anomaly Detection, Severity & Cause Analysis
Agentic AI models to detect anomalies in industrial environments, combining audio, video, and telemetry data. Lightweight small language models (SLMs) perform on-edge analysis, while large language models (LLMs) handle escalation and explanation — enabling fast, safe, and explainable anomaly detection.
Resource Control in Agentic AI Systems
Efficient algorithms for real-time control of shared resources across multimodal AI agents — ensuring fair and optimal allocation for distributed AI workloads.
Intelligent Surveillance System
Developing audio-visual surveillance systems for industrial monitoring and safety that can detect and escalate abnormal human behavior in real-time, leveraging advanced vision and language models.
Technology Stack
The open-source and foundation models behind our research.
PyTorch
TensorFlow
OpenCV
Ollama
LangChain