Uncrewed Traffic Management Team at Wing

• Designed and developed distributed real-time applications for Google X’s autonomous drone delivery initiative, advancing next-generation aviation systems with safety validated via automated end-to-end testing pipelines

• Engineered C++/Scaffolding backend services using Protocol Buffers and S2 Geometry Library

• Migrated geospatial data to Google Cloud Spanner and visualized historical fleet records using an analytical query engine on Alphabet infrastructure, enabling metric monitoring and alerting across critical systems

• Built configuration tools and operator interfaces on Google Maps for annotating and enforcing 4D (3D + time) airspace restrictions, ensuring flight safety and regulatory compliance

• Replaced legacy authentication modules with role-based configurations and low-latency permission checks optimized for critical operations, including pilot-initiated emergency landing commands

• Led development of LLM/VLA-powered multi-drone coordination systems supporting natural language control, intuitive AI-human-robot interaction, 3D simulation for flexible task execution and rapid scenario generation

Wing has successfully graduated from a moonshot project initially under Google X Development to become an independent Alphabet business and launched a first-of-its-kind, on-demand service on three continents, delivering everything from medicine and library books, to hot coffees and fresh cookies to real customers in Australia, Finland, Ireland, the United Kingdom and the United States

Computer Vision Tools Team

• Led end-to-end development of AI-assisted annotation tools for multi-modal data (React/Three.js frontend and Flask/Spring backend), from prototype to production

• Extended point cloud cuboid annotation to semantics segmentation as a side project, later adopted as a new product used by 20+ leading clients in autonomous vehicles and robotics sectors

• Developed a localized platform for the Chinese market as the founding engineer at Appen China

• Explored point cloud object detection/tracking to enhance labelling efficiency using open-source projects

• Conducted sensor fusion experiments using NumPy to project 3D LiDAR points into 2D camera views based on calibration parameters (e.g., position, rotation, focal length) from client devices

• Implemented event loggers, collaborating with UX designers and data scientists to analyze annotator behavior for training process and tools improvement, resulting in a 56% labelling costs

AI WORKS HUE Tech Lead Group

• Built full-stack ERP solutions with Vue.js frontend and Spring/Hibernate backend

• Explored transformer-based chatbot and LSTM-based OCR (Tesseract) for workflows automation

Natural Language Processing Lab

• Researched conversational agents for Chinese short text and maintained large Weibo post–comment dataset

• Implemented Seq2Seq with attention in PyTorch and applied reinforcement learning from human feedback (RLHF) using comment appropriateness scores as rewards