Alignment Interpretability Research mentored by Professor Sharon Li

• Served as a peer mentor for a high level computer science course focused on the design and analysis of efficient algorithms and computational intractability.• Conducted regular office hours, providing valuable guidance and support to students seeking help with course materials and assignments.• Facilitated interactive discussions and study groups to reinforce key concepts and encourage critical thinking among students.

• Responsible for learning and teaching Industry standard software including Adobe Creative Suite and basic programming to students across campus• Delivered open-enrollment workshops and drop-in appointments for students• Prepared and maintained comprehensive documentation for all software and services offered• Engaged in outreach and promotion of the services provided by the Department of Information Technology.

• Worked on a project to utilize explainability techniques on Federated Machine Learning Models.• Implemented FedAvg, FedProx, and FedOpt model aggregation algorithms in Pytorch to analyze the feature attribution of different clients across training rounds and epochs for a CNN model.

• Developed an IOS app to integrate with smart locks over ultra wide band communication to provide indoor AR navigation and access control• Deployed flask based api using docker compose which processes 3d room scans done using lidar scanners and pathfinders through the floorplan using a-star algorithm

• Developed a Python framework to enable machine learning-driven atomic kinetic Monte Carlo simulations for concentrated alloys.• Conducted thorough testing and validation of machine learning-driven calculations, ensuring their reliability and effectiveness in practical simulations.• Applied model selection techniques to enhance computational speed and prediction accuracy, optimizing simulations for production scale applications.

• Developed and rigorously tested NLP large language models such as GPT-3, Bloom, and Bart for data extraction, showcasing a strong proficiency in working with state-of-the-art models.• Utilized advanced libraries and tools including Huggingface Transformers, PyTorch, TensorFlow, OpenAI CLI, and the Python programming language to implement and fine-tune large language models for enhanced data extraction performance.• Applied Parameter-Efficient Fine-Tuning (PEFT) and Supervised Fine-Tuning (SFT) techniques to enable fine-tuning of large language models on extensive datasets, improving their adaptability to real-world applications.• Dockerized training, testing, and inference pipelines to ensure the reproducibility of research work, making them production-ready and further demonstrating a commitment to robust and scalable solutions.• Leveraged clusters from the Center for High Throughput Computing at UW-Madison for large-scale model training and testing.• Collaborated effectively with a diverse team of students, fostering teamwork and honing interpersonal skills while collectively advancing the research project.• Co-authored a research paper proposing a novel method utilizing large language models to automate building materials property databases.

• Conducted research in the multimodal medical domain, focusing on enhancing results through the integration of state-of-the-art models.• Evaluated and tested multiple state-of-the-art models, including BLIP2, Vision Transformers, LSTM Encoder-Decoder, and image feature extraction, to establish baselines for image captioning and VQA on preprocessed X-ray datasets.• Performed data analysis and cleaning, ensuring the quality and reliability of training data, which positively impacted model performance.

• Prototyped a dynamic React-based website for a food chain, showcasing a strong grasp of front-end development and user interface design principles.• Developed a user-friendly graphical user interface (GUI) to monitor cluster computing resources.• Designed and implemented a full-stack chat application using React and Firebase, highlighting proficiency in both front-end and back-end development technologies.

• Member of Wisconsin Racing, a student organization building two cars for Formula SAE international collegiate competition• Part of the firmware team, responsible for writing a library using C and C++ for accessing sensor readings in the battery managementsystem of the electric car• Experienced in C and C++ programming for microcontrollers, utilizing Texas Instruments Tivaware library and UART transmissions

• Helped implement an image segmentation object detection project through performing exploratory data analysis on the dataset• Responsibilities included dataset cleaning and determining the optimal model for regression and classification objectives