AI Prompting with In-Context Learning Explore the power and limitations of In-Context Learning (ICL) with Large Language Models for sentiment analysis. Image Classification & Retrieval with CLIP Zero-shot image classification and text-to-image retrieval using OpenAI's CLIP model, leveraging contrastive image-text embeddings for cross-modal similarity matching. UC Davis Course Advisor A RAG-powered bilingual course advisor for UC Davis students, featuring dual-path retrieval, prerequisite DAGs, and Claude-generated recommendations. Teaching a Small Model to Write Flutter UI Code with RL A pipeline using GRPO + CLIP rewards to improve screenshot-to-Flutter code generation with Qwen3-VL-8B, trained on a single A100 for under $50. Text-to-SQL Generator via QLoRA Fine-Tuning Fine-tuning Qwen2.5-7B-Instruct with QLoRA to generate clean, executable SQL from natural language queries, achieving dramatic parameter efficiency with fewer than 1% trainable parameters. Decoder-only, Shakespeare words generator We build, train, and evaluate a minimal decoder-only Transformer from scratch using PyTorch. We will train this model on the Tiny Shakespeare dataset to generate Shakespeare-like text. data App Version Update — User Analysis A multi-stage statistical analysis of a mobile app version update (S14) across four global markets, examining behavioral shifts, revenue attribution, and user retention. Austin Airbnb Price & Crime Rate Correlation Analyzing Airbnb listing price distributions in Austin, TX, predicting price determinants, and exploring the relationship between neighborhood crime rates and listing prices. Does Classroom Size Affect Academic Achievement? A multilevel statistical analysis of Tennessee's Project STAR dataset, examining the causal effect of small class sizes on student SAT performance across kindergarten through 3rd grade.
