Iman Makaremi is Catio’s co-founding AI leader, building systems that help engineers understand cloud infrastructure and make consequential architectural decisions. His work combines graph-native reasoning, specialized agents, and human evaluation to connect technical dependencies with organizational requirements.
From image recognition to infrastructure intelligence
Makaremi earned a doctorate in electrical engineering from the University of Windsor, researching face recognition in degraded images. He subsequently worked on cloud observability and infrastructure intelligence at Metafor and Splunk, progressing at Splunk from principal data scientist to principal product manager for machine learning and AI. His work on the Splunk Machine Learning Toolkit applied anomaly detection, forecasting, and clustering to operational problems.
After leading data science at MacroHealth, he joined Catio’s co-founding team in 2023 as its AI lead. There, he began developing architectural copilots that reconcile infrastructure topology, business requirements, resource constraints, and competing design choices.
- Compute graphs instead of flattening them. Makaremi found that embedding infrastructure components as text could retrieve relevant services while losing relationships between them. His approach favors graph traversal, centrality, clustering, and shortest-path analysis to calculate dependencies and outage impact directly.
- GraphQA makes structural analysis conversational. His introduction to GraphQA describes an open-source graph-analysis framework that translates natural-language questions into executable NetworkX algorithms. Applications include infrastructure blast radius, influential components, data lineage, and supply chains.
- Specialized agents need explicit coordination. His hierarchical multi-agent architecture assigns responsibilities to a chief architect, domain specialists, and agents retrieving requirements and infrastructure state. Recommendations proceed through generation, conflict resolution, and detailed proposals; isolated conversational histories and cloned specialists enable parallel work while limiting irrelevant context. His writing on multi-agent systems emphasizes structured communication and controlled workflows.
- Human evaluation is part of the architecture. Makaremi’s team built Eagle Eye, an internal interface for examining retrieved requirements, agent interactions, recommendations, and hallucinations. At AI Engineer World’s Fair 2025, he described assessing recommendations for relevance, feasibility, and clarity. More recently, he has outlined agent designs incorporating planning, specialized subagents, persistent working context, and carefully constructed prompts.
