cv
General Information
Full Name | Zhiyi Lai |
Date of Birth | 1996, may, 2nd |
Languages | Chinese(Native), English(Fluent) |
Education
-
2021 Master
UNIVERSITY OF MASSACHUSETTS AMHERST - Master's in Computer Science
- GPA: 3.925 / 4.0
-
2018 Bachelor
HUAZHONG UNIVERSITY OF SCIENCE & TECHNOLOGY - Bachelor of Computer Science and Technology
- Outstanding Graduates
- GPA: 3.63 / 4.0
-
2017 Summer Program
NATIONAL UNIVERSITY OF SINGAPORE - Computer Graphics Summer Program
- Grade: A
Experience
-
2021 - Present Senior Machine Learning Engineer
Xiaomi EV - Build video retrieval system with 500k captioned video data
- Optimized LLaVA-based LMM through SFT/LoRA, empowering Auto-Triage’s problem distribution.
- Build a self-driving multi-modal dataset including 50k video data and 900k image data.
- Enhance the image retrieval system with large detection and segmentation model
- Developed a large detection and segmentation model based on Grounding-DINO.
- Compared the trained model with CLIP-based retrieval system, the accuracy of detecting special obstacles has increased by 50% on average, mined over 600k false and missed detection data.
- Supports 100+ types of special obstacle detection (municipal vehicles, small animals). The entire process of mining new categories is completed within one week.
- Developed a CLIP-based image retrieval system
- Maintained a vector database of 7.5 million frames and about 1 million kilometers of data with CLIP.
- Improved the road condition and single target search results by combining the detection results.
- Build video retrieval system with 500k captioned video data
Projects
-
2024-now LOASISIndependent Developer
- A game project based on LLM, using LLM as the host to generate DnD-Like game stories.
- Adjust the generated story by Prompt Engineering to make it fulfill the game’s theme and length limit.
- Use Streamlit to build the front-end, and use MongoDB to store game and user data.
- A game project based on LLM, using LLM as the host to generate DnD-Like game stories.
-
2024-now Awesome AI in GameOpen Source Contributor
- An open source page that organizes game-related AI papers and projects. Including action generation, RL and LLM-related game papers as well as open source game environments, LLM games and more.
-
2024-now CAMELOpen Source Contributor
- CAMEL is an open-source library designed for the study of autonomous and communicative agents.
- Implemented Azure OpenAI API support for CAMEL.
- CAMEL is an open-source library designed for the study of autonomous and communicative agents.
Academic Interests
-
AI in Games
- LLM generated game content
- PCG (Programmatic Content Generation)
- Animation
- 3D Generation
Other Interests
- FPS Games: Valorant, CS2, Overwatch, etc.