Faster Coding Agents with Hypercode | OpenAI Codex Hackathon Winner Demo

HyperCode replaces grep-based search with context-aware file retrieval for coding agents, reducing latency, token usage, and missed dependencies. 👉 See the demo 👈 In this video, Maks Operlejn and Szymon Janowski, Senior ML Engineers at deepsense.ai, present HyperCode — a repository intelligence layer that won the @OpenAI Codex Hackathon in San Francisco. HyperCode sits between a code repository and a coding agent. Instead of iterative search, it provides ranked, context-aware groups of files at the start of execution. Key ideas shown in the demo: 🔍 replacing grep-style search with structured context retrieval 🔍 ranking files by relevance to a natural language query 🔍 capturing Codex session memory and decisions 🔍 building a hypergraph of code structure, commits, and semantic labels 🔍 reducing iteration loops in coding agents Results observed during testing: 🎯 over 50% faster plugin execution 🎯 reduced token usage 🎯 improved recall of relevant files 🎯 fewer missed dependencies across the codebase #AIAgents #AIEngineering #OpenAI #DeveloperTools #LLM #SoftwareEngineering #codex #hackathon #openaicodex If you have questions for Maks and Szymon, feel free to comment below. 💭 🔗 Check out our website: https://deepsense.ai/ 🔗 Linkedin: https://www.linkedin.com/showcase/applied-ai-insider