How an Applied AI Engineer Used CreateOS to Deploy an AI Agent Benchmarking Game in Under an Hour

• —Two AI agents pitted against each other in a live puzzle maze environment, observable in real time
• —Built and deployed to a working URL during a spontaneous conversation at the AI Impact Summit, Delhi
• —Entire flow completed before a coffee order could be fulfilled
• —A lifelong coder's deployment problem solved in a single session, without infrastructure overhead

Pratyush is an Applied AI Engineer at AGI House India who has been writing code since the age of six. He works at the intersection of applied machine learning and agent research, and is technically fluent across the full development stack. When the NodeOps team found him at the AI Impact Summit in Delhi, he was mid-session trying to figure out how and where to deploy something he had already built. The problem was not the idea and not the code. It was the deployment.

Pratyush wanted to build an environment where two large language models could be placed against each other in a puzzle maze and observed navigating it. The research value is in watching how different models reason through complex, constrained environments: which paths they choose, where they get stuck, how they adapt. The concept was clear and the technical instincts were there, but translating the idea into a deployed, shareable application meant solving an infrastructure problem that had nothing to do with the experiment itself.

Deployment is often where business momentum dies. For builders who think in models and agents rather than in servers and configuration, the gap between a working concept and something live and accessible represents a context switch that costs real time. Pratyush was working through that gap manually, evaluating options, when the conversation started.

The business need was specific: get a working AI environment deployed and accessible without the overhead of managing infrastructure manually. CreateOS solved the deployment problem directly. The builder described the application, an arena where two AI agents navigate a puzzle maze, and CreateOS used AI to handle the infrastructure reasoning, generate a working version, and prepare it for deployment, all within a single conversation.

No server configuration. No hosting decisions. The AI-driven build process took the concept from description to deployed application through GitHub, producing a live URL for Agent Blaster before the session had time to conclude. The deployed product is now actively in use as a research tool, giving the builder a live environment to observe how different AI models behave under constraint. The deployment barrier that had been blocking progress was removed entirely.