“Mingqi” Medical Imaging Large Model Matrix

Subproject Leader Dec 2024 – Present

• Multi-granularity features + domain-specific LLM → endoscopic reports with >20% higher semantic similarity.
• Enhanced cross-modal image-text retrieval via fine-tuned PMC-CLIP, achieving precise case matching with >30% higher accuracy.
• Built 20-hr Mandarin medical speech corpus (via voice cloning), reducing ASR false positive rate by >20%.
• Recognized as one of the Top 10 AI Application Scenarios at the 2nd Belt and Road Sci-Tech Exchange Conference; awarded Top 10 Outstanding Cases of AI for Sustainable Development.

ArXivToday-Lark: Automated arXiv Paper Delivery for Lark Teams

Key Team Member Mar 2025 – Mar 2025

• Developed a lightweight automation tool that fetches the latest arXiv papers daily and delivers them directly to Lark group chats via a custom bot, streamlining research discovery for academic teams.
• Features include scheduled fetching, customizable topic filters, LLM-powered paper summarization and translation, and influence prediction (in development).
• Designed for extensibility—supports integration with other messaging platforms and backend enhancements for intelligent paper recommendation.

T2I-Fuzzer: Jailbreaking Text-to-Image Large Models via Hybrid Strategies

Key Team Member Sep 2024 – Oct 2024

• Designed multi-level attack strategies to expose vulnerabilities in T2I models, optimizing adversarial efficacy from a zero-sum game perspective.
• Implemented a unified attack and evaluation framework, enabling efficient and stealthy rewriting of harmful prompts to bypass safety filters in T2I systems.
• Participated in the overall framework design and conducted testing of attack schemes such as adversarial captioning; built an attack knowledge base and implemented automatic prompt sanitization.

Web3_MLM: Intelligent Reverse Analysis for Solidity Smart Contracts

Team Member Jan 2024 – Apr 2024

• Supervisor: Chao Zhang, Associate Professor, Tenured, Deputy Director of Institute for Network Sciences, Tsinghua University
• Web3_MLM is a subset of the MLM project, providing an intelligent reverse analysis solution for Solidity smart contracts.
• Constructed a high-quality, comprehensive dataset containing Solidity source code, assembly code, and bytecode; performed data augmentation and aligned data formats at both contract-level and function-level granularity.

RMDB: Database Management System for the National Contest on System Capacity

Key Team Member May 2024 – Aug 2024

• Enhanced a database management system within a given framework to support MySQL-like SQL functionalities.
• Implemented storage management module, contributing to over 35% of the team's completed tasks in the preliminary round.
• Participated in optimizing the COUNT(*) etc. query, achieving approximately 10% performance improvement.

SysY-to-ARM Compiler

Team Header Oct 2023 – Jan 2024

• Developed a compiler from SysY (educational language) to ARM assembly in C++, covering full compilation pipeline.
• Implemented lexing (Flex), parsing (Yacc), backpatching-based control flow translation, and optimizations (CSE, DCE).

Python Code Completion System Based on Neo4j Graph Database

Team Member Jul 2023 – Aug 2023

• Developed a context-aware Python code completion system using Neo4j as a graph database to infer variable types, addressing the lack of built-in type hinting in standard Python interpreters.
• Responsible for syntactic analysis of input code and querying the graph database; implemented path traversal logic to retrieve successor nodes, perform type inference, and return results to the frontend.
• Designed with extensibility in mind—can be adapted into a VS Code extension for real-time intelligent code suggestions.

JingZhiYao: A Blockchain-Based Scheme for Precise Rumor Refutation and Traceability

Team Member May 2023 – Jul 2023

• Designed a cross-platform rumor governance scheme leveraging blockchain technology, aiming to optimize key stages in rumor management including rumor identification, precise refutation, and traceability for accountability.
• Responsible for the design and implementation of the database, as well as backend database integration; achieved secure data upload, encryption, storage, and smart contract-based retrieval for message matching, verification, and filtering, supporting information traceability and evidence preservation.

Zouma: A C++ Graphical Game with Winter Olympics Theme

Course Project Apr 2022 – May 2022

• Developed a Zouma-style puzzle game in C++ using the EasyX graphics library, featuring a custom game loop, path-based ball movement, and interactive cannon control with real-time mouse tracking.
• Implemented core mechanics including ball trajectory generation (via linear interpolation of key points), collision detection, ball elimination logic, and dynamic visual effects (e.g., rotating "Bing Dwen Dwen" character).
• Integrated background music and sound effects using Windows MCI API; designed start screen, win/loss detection, and smooth transitions between game states.
• Awarded **2nd place in class** for creativity, code quality, and user experience in the course project evaluation.