Chicago, Illinois, United States
STAT 24300 Numerical Linear Algebra, Department of Statistics
Developed the EV trip simulator and data generator using first principles. Generated realistic telemetry data (time vs. position, power, battery level, weather, etc.) for training and testing energy consumption predictor, mitigating out group’s limited access to real-world EV trip data. Developed the seq2seq transformer EV energy consumption predictor based on TensorFlow with training and inference pipeline for reducing charging costs by optimizing charging schedule. Developed the Gaussian convolutional encoder to capture the spatial dependency of route data. Integrated into the linear regression model for interpretable prediction of energy consumption. Tracked tasks and completed sprints with Jira to quick-fix bugs in models and pipelines. Managed feature branches of models and pipelines in the company’s monorepo using Git.
Project: Multi-Activation Function Neural Network for Function Learning in Physical Laws Proposed and developed the multi-activations functions neural network for learning mathematcial functions from data to enhance interpretability and generalizability in predicting physical variables. Implemented the novel neural network with training and inference pipeline using TensorFlow. Developed a Newtonian motion simulator to generate data for testing the neural network.
Project: YIZHE: An AI-Facilitated Post-clinic Service Chatbot Used OpenAI API and prompt engineering to develop the GPT-4o-based post-clinic service chatbot. Deployed in user experiments with a real clinic’s customer service group chat. Designed the instructions, response template, and few-shot training data generator with state-of-the-art methodologies in prompt engineering for medical agents, including few-shots learning, discrete prompting, and prompt generation. Co-authored a 19-page paper using LaTeX.
I implemented a long-short term memory (LSTM) neural network with training and inference pipeline in TensorFlow for predicting stock prices in equity investment. Tested model with CSI 1000 and Microcap 500 indices stocks. Visualized the performance macroeconomic indicators for stock-index futures investment using Python (NumPy, Matplotlib...) to provide a straightforward illustration for decision-making of fund managers. Developed a stock data crawler using OpenCV and PyAutoGUI that saves stock data locally for model training and testing, mitigating our team’s limited access to data API.