Zhengyuan (Dora) Dong

Status: Completed ✅ at Jun 2025. Updated at Dec 2025

ModelTables is a benchmark corpus of tables in Model Lakes that captures structured semantics of performance and configuration tables often overlooked by text-only retrieval. Built from Hugging Face model cards, GitHub READMEs, and referenced papers, it links tables to their surrounding model and publication context. The corpus covers over 60K models and 90K tables, with multi-source ground truth using citation links, model-card inheritance, and shared training datasets. We evaluate table search methods including Data Lake operators (unionable, joinable, keyword) and IR baselines (dense, sparse, hybrid retrieval), demonstrating the first large-scale benchmark for structured model knowledge discovery.

Status: Completed ✅ at Mar 2025. To Be Released

LazyVLM is a neuro-symbolic video analytics system that combines the flexibility of Vision Language Models (VLMs) with the efficiency of symbolic methods. It allows users to query open-domain video data at scale using a semi-structured text interface, decomposing complex video queries into efficient operations for robust and scalable analytics.

Status: Completed ✅ at Oct 2023. Updated at Oct 2024

An AI-driven chatbot platform that simplifies bioinformatics data analysis through conversation. Features include front-end and back-end components, extensive data setup, model fine-tuning, and deployment solutions across Docker, Railway, and terminal CLI.

Status: Completed ✅ in Jun 2023

A GUI application that downloads and manages GitHub repository README files locally while offering integrated web search functionality through popular search engines. The tool streamlines documentation access by automatically fetching README files from repositories and displaying them in a user-friendly interface for offline browsing.