About Me
Experience
Machine Learning Engineer
Unknown - - Now
职位: Machine Learning Engineer | 时间段: Unknown | 工作内容: Developed a financial data processing module independent of LLMs for financial data computation tasks. Currently developing a financial LLM for tasks such as summarization, analysis, and documents writing. Achieved financial data processing automation, which is 1000x faster than manual processing. Constructing a financial text vector-database to fine-tune and apply RAG to the base LLM.
Machine Learning Engineer
Unknown - - Now
职位: Machine Learning Engineer | 时间段: Unknown | 工作内容: Built an improved Conv-LSTM model using Pytorch for nationwide, high accuracy air pollution forecasting, achieving 100x faster inference speeds than traditional models. Accelerated training process by converting data from 64-bit double precision to 16-bit half precision using Apex tool. Reduced training time per epoch from 70 minutes to 20 minutes using Pytorch's pipeline tech.
Head of Business Operations
SJTU-WALKER - - Now
职位: Head of Business Operations | 时间段: Unknown | 工作内容: Responsible for business planning, financial analysis, and pitch presentations. Coordinated with incubators, led the team to win over 10 innovation competition awards and received more than ¥100,000 funding. Communicated with Stanford University, WeRide, Tsingyuan Ventures in San Francisco, US on behalf of the team.
Development Team Member
SJTU-WALKER - - Now
职位: Development Team Member | 时间段: Unknown | 工作内容: Deployed the YOLO object detection algorithm on Nvidia Jetson TX2 platform and developed lane detection algorithm based on SCNN.
Machine Learning Engineer
Unknown - - Now
职位: Machine Learning Engineer | 时间段: Unknown | 工作内容: Implemented AI for auto defect detection to improve efficiency and accuracy while reducing costs. Utilized lightweight SqueezeNet to improve inference speed and employed pre-trained models to improve accuracy. Experimented with Neural Architecture Search (NAS) to find a more effective network. Increased inference speed from 2-3 fps to 24 fps and accuracy from 93% to 98.5%.