I build AI agents, graphics systems, and desktop AI applications. Currently focused on AILIS.
- HPG 2026 Competition Champion: Procedural Neon Metropolis GI - ShaderToy Demo
- Nowcoder AI Coding Challenge Cup Champion
- Tencent Game Algorithm Competition - High-Performance AI Rendering, Top 9 (Solo): developed an AI lightmap compression solution and integrated it into an Unreal Engine rendering pipeline, reducing scene texture data from 40 GB to 250 MB while targeting a stable 60 FPS on an RTX 4060.
AILIS is an open-source desktop AI companion with an embodied VRM character. It is designed to stay on the desktop, learn about its user, handle everyday tasks, and complete more complex work instead of being limited to chat or desktop-pet interactions.
- Embodied desktop character: built with Electron, Three.js, and VRM, with expressions, gestures, voice interaction, MToon rendering, and installable character packs.
- Task execution: understands what is on screen, reads files, researches information, reviews code, analyzes problems, and uses tools to complete tasks.
- Humanized responses: separates task execution from character presentation, turning tool calls, logs, and raw results into natural responses.
- Long-term memory: remembers user preferences, project context, and output habits to provide more personalized assistance over time.
Repository: github.com/haowenGuo/AILIS
C++ Python TypeScript Electron Three.js VRM Unreal Engine FastAPI SQLite Chroma Ollama vLLM