Alisha Li

AI-Driven Ideation R&D: Led the development of AI tools for ideation by combining random word generation with trend analysis to drive innovative product development. Prompt Engineering for Research: Created a prompt chain system using ChatGPT and LLMs to automate white paper research, improving research efficiency for R&D projects. AI-Driven Synthetic Interviews: Developed and implemented an AI-powered tool leveraging auto agents and the ChatGPT API to simulate UX research with virtual personas. Automated Web Scraping Pipelines: Designed automated pipelines to extract product lists, news trends, and competitor insights from websites for real-time analysis. NLP and Machine Learning in Interview Analysis: Applied NLP and machine learning techniques (K-Means, Hierarchical Clustering, topic modeling) to analyze interview scripts, including sentiment analysis using TextBlob. Tools: Python, Scikit-learn, BERT, Word2Vec, Gensim, NLTK.

Machine Learning Pipeline for Roof and Building Feature Detection Developed a machine learning pipeline to detect roofing materials, roof condition, number of roof tiles, as well as the presence of doors and windows using YOLOv8 and InceptionV3 on a dataset of over 1,000 labeled images. Trained a YOLOv8 model on 137 images for roof base state detection, using data augmentation techniques (rotation, shifting, zooming) to improve robustness. Applied transfer learning with InceptionV3 to classify five types of roofing materials, achieving 96.02% training accuracy and 75.81% validation accuracy. Built a roof tile state detection model using a custom architecture based on InceptionV3, incorporating early stopping and the Adam optimizer. Data augmentation was used to improve classification stability. Tools & Libraries: Python, TensorFlow, Keras, YOLOv8, InceptionV3, ResNet50, OpenCV, Image Analysis, Computer Vision Models

Provided daily assistance for lectures and kept close contact with students.

Assisted in the sampling survey of the Economic Statistics Project at Zhengzhou, distributed questionnaires, collected, and sorted sample data. Reported sample’s abnormal situation, screened invalid data and optimized obtained data according to initial analysis results and actual information.