● Led a 10-member team in developing an adaptive AI math tutoring agent with self-assessment–driven feedback loops, leveraging LangChain for multi-agent orchestration and a Django backend; achieved ~18% improvement in student post-test performance, demonstrating measurable learning gains beyond standard LLM chatbots.

● Built on LangChain with a dual-backend architecture (GPT-4.1 + Gemini-1.5-flash) and a ReAct Loop evaluation pipeline, the system enables structured, multi-dimensional reasoning and adaptive tutoring while achieving 99% uptime, reducing hallucinations by 25%, and maintaining an average response latency of under 2.1 seconds even under high concurrency.

● Implemented a dynamic weight-adjustment feedback loop that updates student profiles after each query, improving personalization by 30% and reducing redundant explanations by 20%

● Instruct 150+ students in React.js, Bootstrap, JavaScript, and web development techniques; facilitate their use ofndevelopment environments (Glitch, VSCode, IntelliJ IDEA) and GitHub for version control

● Assist in grading assignments and evaluating over 80+ web application projects, including those involving API integration, backend database setup, and authentication SDKs

● Initiate a comprehensive review of course materials, correcting code errors and enhancing teaching quality

● Contributed to the development of a DeepSeek-based address agent embedded in the mobile app via WebView, enabling field engineers to accurately locate telecom base stations across 50+ Chinese cities with ~93% query resolution accuracy in internal testing.

● Developed a fault-tolerant markdown streaming parser with regex-based sanitization, reducing malformed LLM output rate from 18% to 3% in benchmark tests and lowering effective user-perceived latency by ~30%.

● Developed an integrated system for managing communication data and tenant information, reducing

management costs by 20% and improving efficiency by 25%

● Optimized Excel data processing by replacing the Apache POI library with EasyExcel, reducing memory consumption by 90% and increasing data processing speed by 80%

● Implemented Screen Capture API for web page recording, replacing third-party plugins and improving

efficiency by 15%; implemented robust data validation in backend code, ensuring 100% compliance with data standards and seamless database integration.

● Implemented Continuous Integration/Continuous Deployment (CI/CD) pipelines with Git that accelerated deployment cycles by 20%, enabling more frequent feature rollouts

● Integrated 3000+ data from 4 research projects by R-studio and Stata to ensure the accuracy of datasets

● Contributed to data tabulation and literature review editing for 10+ research papers to follow research guidelines

● Contributed to early-stage development testing of ScratchJr software, offering 20+ feedback for enhancements