Yun-Hao Lee

About

Robotics/AI candidate who built an AI-driven landmark location tool during a summer internship at Mobile Drive, combining YOLOv5 object detection with OpenStreetMap-based geolocation to handle dense, cluttered urban environments. Also researched deploying LLM-based agents on constrained hardware using quantization plus LoRA/continuous learning, improving accuracy from ~80% to ~92%, with an emphasis on production logging for reliability.

Experience

Education

Key Strengths

• Improved YOLOv5 object detection accuracy by 45% for robotic vision landmark detection
• Designed hybrid perception approach combining OpenStreetMap API (long-range) with YOLOv5 bounding boxes (short-range) for dense urban environments
• Optimized AI agents for hardware-restricted real-time environments using quantization
• Raised model accuracy from ~80% to ~92% via LoRA fine-tuning and continuous learning
• Production reliability mindset via comprehensive logging of prompts, responses, and user feedback for debugging
• Built and deployed an on-device, privacy-focused LLM agent with hardware-aware model recommendations
• Improved model accuracy by ~37% while reducing memory usage
• Strong customer alignment via weekly check-ins and clear translation of complex technical concepts
• Rapid production debugging under pressure (patched missing logprobs issue and restored live demo without downtime)
• Pragmatic approach to prototyping vs. production code with planned refactoring