Erina Karati

Erina Karati builds applied AI systems spanning generative AI, multi-agent architectures and production-ready machine learning pipelines. Her 2026 conference biography describes her recent work as an AI Engineer at Supercell, building modular multi-agent systems and scalable AI infrastructure for interactive environments, with resulting research accepted to the WiML Symposium at ICML 2026.

At Supercell’s AI Innovation Lab, she developed Project Paradox with Arunachalam Manikandan: a modular framework for autonomous game agents that interact, compete and cooperate with players and one another. The framework combines separate retrieval-backed memory namespaces, emotion vectors, trust scores and memory importance scoring, enabling agents to plan actions, manipulate objects and hold conversations that influence their beliefs and goals. In her presentation on long-horizon agents, Karati examines how extended interactions can erode source attribution, turn rumors into apparent facts and disconnect remembered information from planning. She presents an autoresearch loop that runs controlled scenarios, collects structured traces and evaluates whole-run behavior across information reach, source retention, uncertainty preservation, replanning and privacy containment. With the evaluation harness, scenarios and metrics held fixed, the loop tests small changes to memory, communication, trust and planning policies, retaining changes only when the scorecard improves and guardrails hold. She reports a more contextual response in a mango-rumor example while reserving broader improvement claims pending repeated results.

Previously, Karati spent three years at Microsoft working on large-scale production systems in complex enterprise environments. Her work covered networking, system reliability, security and debugging distributed failures at global scale, informing her approach to robustness, observability and reliability in AI systems. She is also the co-founder of MinneDigest, an AI-powered news and podcast platform that won the AI x Journalism Hackathon and secured $10,000 in grant funding. She graduated from the University of Minnesota with a master’s degree in computer science and a 4.0 GPA in May 2026. Her interests center on combining strong engineering foundations with advanced AI to build meaningful systems for real-world use.