Arunachalam Manikandan

AI Engineer, Co-Founder

University of Minnesota

Arunachalam Manikandan is a Computer Science graduate student at the University of Minnesota, specializing in large language models, deep learning, vision-language systems, retrieval-augmented generation (RAG), and agentic AI. His work focuses on translating cutting-edge research into scalable, real-world AI products. He currently serves as a Graduate Research Assistant, where he develops advanced deep learning pipelines for biomedical image segmentation. His research involves applying large vision models such as Segment Anything Model (SAM) and Vision Transformers (ViT), enhanced with custom adapter layers, to perform whole-slide image segmentation on complex histology datasets. His work spans end-to-end system design, including multi-stage distributed training on high-performance computing clusters, patch-based inference, and structured outputs for downstream analytical workflows. Previously, Arunachalam was part of Supercell’s highly selective AI Innovation Lab in Helsinki, where he co-developed Project Paradox, a novel multi-agent framework. This system integrates agent memory, personality, emotional modeling, and dynamic planning to create emergent, story-driven gameplay experiences. His work was presented to senior leadership, including company executives, demonstrating the potential of agentic AI in interactive entertainment. In addition to his research and industry experience, Arunachalam has contributed to multiple early-stage startups, helping take AI-driven products from concept to deployment. He is also a frequent hackathon participant and co-creator of MinneDigest, an AI-powered newsroom platform that won the AI + Local News Hackathon and secured a $10,000 grant to support continued development. Earlier in his career, he worked as a Software Development Engineering Intern at Visa Inc., where he built scalable Spark and Scala-based data pipelines for processing high-volume financial transactions. Arunachalam is passionate about building intelligent systems that combine reasoning, perception, and real-world impact, and is actively pursuing opportunities in AI engineering and applied research.

Sessions (1)

Autoresearch in a Multi-Agent AI Village

3:45 PM·Main Stage

Project Paradox is an existing multi-agent framework built at Supercell's first AI Innovation Lab, which has a 3D Unity village with local LLM powered agents. The characters remember conversations, update emotional state, track trust, plan actions, move through rooms, transfer items, and talk to each other through a FastAPI backend. The new work is an autoresearch layer around that village. We built a backend loop that runs controlled social scenarios, scores the resulting NPC behavior, proposes protocol or policy changes, reruns the suite, and keeps changes that improve the agents. The goal is to move beyond one good chat response and measure whether an NPC society can preserve source attribution, verify claims, spread important information, coordinate goals, and replan after new information arrives. The talk walks through the system architecture and the lessons from building it. We show the backend simulation harness that executes Unity style actions without opening Unity, the scenario suites that test information diffusion and memory provenance, and the ratchet loop that edits protocol text or planner policy with rollback. One accepted run improved information diffusion by teaching agents to broadcast important sourced evidence while preserving who said it. The practical takeaway is a reusable pattern for AI engineers building agents with messy state. Freeze the harness, expose a small editable policy surface, score real behavior instead of vibes, and let an agent search for improvements under rollback. The same pattern applies to game agents, coding agents, support agents, personal agents, and other systems where long horizon behavior matters more than a single response. Speakers: Erina Karati — Former Microsoft, Supercell; Arunachalam Manikandan — University of Minnesota.