4+ years of product experience across large-scale platform products at Meituan and 0-to-1 AI products in startup environments. Experienced in LLM, Agent, RAG, OCR, CV, and robotics-based AI applications, with full-cycle ownership from product planning, user research, requirement definition, solution design, launch execution, to post-launch evaluation. Experienced in full-lifecycle B-end and C-end AI projects covering transaction supply, user experience, and efficiency improvement.
Responsible for building AI-driven smart platforms, AI operation assistant, and AI-assisted teaching tools. Skilled in Figma, Vibecoding (codex, claude code); Languages: English (CET-6): 530, Mandarin (Native).
C12.ai (AI+Lab Robotics)
Responsible for building the Talos laboratory robot software product from 0 to 1. Assisted chemists in executing experiments (TLC, column chromatography, LC-MS, component collection, rotary evaporation, weighing) via a bilingual smart conversational platform and automated commands. Architected a self-learning agent to help chemists accumulate company "skills" and improve experimental efficiency.
Fiture (AI Fitness Mirror)
AI+Live Training Department (Software-Hardware + B2B / B2C)
Led coach-side products for the online AI+ live personal training business. Built a coach daily operation app from 0 to 1 and collaborated with the PC product team to develop AI-assisted teaching, coaching supervision, and operation tools.
Meituan
Shanghai
Master
Bachelor
GPA: 3.5/4.0 (Top 3%); First Prize Scholarship (1.8%); 2018 Merit Student of University (5%)
Background: In daily experimental workflows, chemists face many repetitive and tedious waiting processes. Accumulated experimental expertise can only be manually transferred and cannot be reused or precipitated as corporate assets. Especially for pharmaceutical companies where launching a drug one day earlier can generate an additional $1 million, improving chemists’ efficiency to focus on complex new drug R&D challenges is critical for boosting ROI. Measures: Architected an agent platform integrated with VLA (Vision-Language-Action) robots to assist chemists in issuing experimental commands and verifying results. Accumulated effective experiential “skills” to feedback into the platform and company, thereby improving AI recommendation accuracy. Responsibilities: Built the ‘Talos’ smart laboratory platform, enabling chemists to query basic chemistry knowledge, issue parameters to robots for experiments, and confirm experimental results. Designed AI capabilities to assist in recommending parameters, providing preliminary judgments on execution results, and recommending next steps, helping chemists improve practical execution efficiency. Facilitated the business side in accumulating chemists’ experiential ‘skills’ into corporate assets. Achievements: Execution Efficiency: Projected to increase chemists' execution efficiency by 6-12x (from 1-2 hours to 10 mins). Successfully ran the POC for the column chromatography scenario within 1 month, and accumulated 50 regularly triggered skills within 2-3 months.
Background: Currently, coaches lack personalized guidance in private classes and summary guidance for common issues in group classes, relying only on basic guidance and some AI teaching voices to improve students' exercise performance. There are also no tools to help junior coaches raise their teaching quality above average. Measures: Optimized fitness industry vertical models based on students' historical training data, input common problem guidance and the progression/regression movements of each movement in the movement library into the RAG database. Identified student’s incorrect posture and give the professional guidance suggestions to the coach and even conclude the same issues of group classes students through TTS when necessary. Responsibilities: Sorted out common issues and guidance from 2,200+ movement libraries, built a RAG database for training problems, selected 200+ high-priority movements, collaborated with teaching teams to collect students' training data, and trained the in-house fine-tuned LLM Qwen2.5 with 20-100 videos per movement to ensure data accuracy. Combined OpenCV skeleton recognition with LLM to output guidance on teaching prompts, improving junior coaches' teaching skills. Achievements: Expected to improve the teaching quality of 86% of normal coaches on the platform, increase user experience by over 1,300 class hours/month, and reduce the platform's average training hours by 60%.
Background: Hosts' daily operations lacking of data-driven guidance for refined their operations will bring negative experiences like guests arriving with no room available, which in turn reduces the transaction fulfillment rate. Measures: By integrating operational data, work orders, and a knowledge base, we've created an AI assistant to provide personalized guidance. Responsibilities: Analyzed over 1 million host service tickets from January to June, identifying and summarizing 3 major categories of frequent issues. It was determined that AI could cover 41% of operational problems, resolving about 30%, with the first phase solving 12.3%. Solutions for issues such as hosts not accepting reservations, being unreachable, or guests arriving to find no room available were found through intent recognition, platform-mediated negotiations, and property recommendations. Benefits: Expected to reduce host consult labor costs by 140k, recover 156 dropped orders/day, and add 328 new orders/day.