Help & Supportcareers@gravityer.com
Senior ML/LLM Engineer | Post-Training Infrastructure, RL/Agent Environments, Inference Optimization & Data | Amazon
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
I’m fortunate to work across the full lifecycle of training 400B+ MoE foundation models from scratch, including pre-training data processing, post-training infrastructure, evaluation, inference optimization, and production-scale agents. I’ve built LLM systems for Rufus, Amazon Ads, and Alexa+, including SFT/RL infrastructure, vLLM/SGLang serving and optimization, evaluation systems, context engine, and tool-use agents in VM-based environments.
Carnegie Mellon University
Master’s Degree, Software Engineering and Management
January 1, 2016 – January 1, 2017
Huazhong University of Science and Technology
Bachelor's degree, Electrical Information and Communications
January 1, 2012 – January 1, 2016
Amazon
Sr. ML/LLM Engineer, Ads Foundational Modeling
October 1, 2025 – Present
Amazon
Sr. ML/LLM Engineer, Stores Foundational AI
June 1, 2024 – October 1, 2025
Amazon
Sr. MLE, Smart Home Machine Learning
July 1, 2022 – June 1, 2024
Amazon
MLE II
October 1, 2019 – June 1, 2022
Amazon
MLE
February 1, 2018 – September 1, 2019
Amazon
Software Engineer Intern
May 1, 2017 – August 1, 2017
Greater Seattle Area
Carnegie Mellon University
Research Assistant - NASA Jet Propulsion Laboratory Collaboration
September 1, 2016 – August 1, 2017
San Francisco Bay Area
Uber
Software Engineer Intern & HRBP Intern
May 1, 2016 – July 1, 2016
Wuhan, Hubei, China
Smart in Home, startup
Software Engineer Intern
November 1, 2015 – April 1, 2016
Wuhan, China
Huazhong University of Science and Technology
Research Assistant @ Media and Communication Lab
October 1, 2015 – March 1, 2016
Wuhan, China
Sicun Studio
Studio Head
March 1, 2015 – June 1, 2016
Wuhan, China
DJI
Robot Vision Developer
June 1, 2014 – September 1, 2015
Wuhan, China
Bingyan Studio
Software Engineer
March 1, 2014 – September 1, 2015
Wuhan, China
QuickBooks Small Business Hackthon in Silicon Valley
October 1, 2016 – Present
In the hackthon competition held by QuickBook in Silicon Valley on October 23th, our group "Jido Donor" used Jibo connecting with Google api.pi, Square, Postman, and QuickBooks API to finished a complete Jobo Donor, making donation process fun. We finally won over other 60 teams and got the Jibo prize.
Autoinhome Smart Home Product Online Store
March 1, 2016 – Present
- Implemented the online store with Angular.js which showed great performance for one-page web app - Accomplished the responsive design to make it look great on mobile devices
VALSE 2016 Computer Vision Conference Website
February 1, 2016 – Present
- Web full stack, wrote frontend with HTML/CSS/JS/bootstrap and back-end program with Python - Designed conference websites and made branding for VALSE 2016 - Opened resource, the website template favored and used by CCBR, Chinese Conference on Biometric Recognition
IEEE SIDAS 2016 International Conference Website
February 1, 2016 – Present
- Web full stack, wrote frontend with HTML/CSS/JS/bootstrap and back-end program with Python - Designed conference websites and made branding for IEEE SIDAS 2016 - Attracted more than 2,000 registered users and over 280,000 page views in two months
End to End Trainable Speech Recognition System
January 1, 2016 – June 1, 2016
- Constructed a 16-layer end-to-end trainable deep learning networks cascading CNN layers, RNN layers, and a CTC layer, based on CRNN architecture using Lua, C++, and Matlab - Proposed to harness original spectrogram pictures instead of MFCC features as CNN input to keep the originality of audio, boosting 9.2% via comparative experiments - Achieved phone error rate 19.65% on TIMIT dataset, competitive to the state of the art result 16.5%
Zuiwant Media Platform
January 1, 2016 – March 1, 2016
- Proposed the idea of changing the hierarchy multi-media info into flat. This idea attracted Tencent QQ Browser product manager for future collaboration. - Lead the team as product manager and implemented rapid development in two months. - Product video: https://vimeo.com/165129244.
Autoinhome Company Website
November 1, 2015 – January 1, 2016
- Implemented the website with Fullpage.js - Accomplished the responsive design to make it look good on mobile devices
Autoinhome Smart Home Product Recommend System
February 1, 2015 – Present
- A common confuse is that you want to make home smarter, but not know how. This recommend system introduces smart home products just after several quick questions and explores your interested products. - Accomplished the responsive design to make it look great on mobile devices
Human Capital Management Network Model
December 1, 2014 – January 1, 2015
- Human Capital Management is an inter-discipline problem. Six problems challenging the ICM organization are solved by proposing a human capital network. - Proposed the multi-layer networks model consisted of trust network, information network, and relationship network, built by three steps: deciding nodes, making bold assumptions of allocation and connection, and determining all edges. - Implemented the revised MCG Matrix with Matlab, to evaluate the impact of churn, balance hiring budget, and maximize the personnel work efficiency in the long run. - Won the Honorable Mention of ICM, Interdisciplinary Contest In Modeling: http://www.comap-math.com/mcm/2015Certs/39879.pdf
BIOMOD HUST-China 2014 Website
August 1, 2014 – October 1, 2014
- BIOMOD is an annual biomolecular design competition for students. - Co-designed website with Photoshop, Illustrator - Implemented website with HTML/CSS/JS/jQuery/FullPage.js/ScrollJS - Won Team Silver Prize: http://biomod.net/winners/2014.html. - Mentored BIOMOD 2015 Website: http://hustchina2015.github.io/. The team won Gold Prize and ranked 3rd among all the teams: http://biomod.net/winners/2015.html.
Robert van Embricqs Parametric Furniture Design System
June 1, 2014 – August 1, 2014
- Constructed Parametric Design System to generating the Robert Van Embricqs rising side tables. With variable parameters, tables with distinct looks rapidly generated via Matlab GUI. - Implemented multiple objects programming with LINGO to maximize the stability of furniture, simplification of manufacture and conservation of materials. - Designed table edge discretization mathematical model. - National First Prize in 2014 Chinese Undergraduate Mathematical Modeling Contest in Modeling: http://dwz.cn/3Hm0Co
AWS Certified Solutions Architect – Associate
Amazon Web Services (AWS)
June 24, 2026 – Present
TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning
Coursera
June 24, 2026 – Present
Cultural Fit Analysis
The candidate's project portfolio shows a diverse range of interests from web development to mathematical modeling and deep learning, indicating a broad intellectual curiosity. Their involvement in hackathons and student organizations suggests a collaborative spirit and a willingness to engage in diverse projects. However, the target role is 'Data Analyst' while the candidate's experience is heavily skewed towards 'Sr. ML/LLM Engineer' and 'MLE'. This significant mismatch in role focus suggests a potential cultural fit challenge if the candidate is not genuinely interested in a pure data analysis role, which typically involves less direct model building and more data interpretation, reporting, and business intelligence. While they have strong data processing skills, the analytical depth for a data analyst role, beyond ML feature engineering, is not explicitly demonstrated.
Soft Skills & Operational Fit
The candidate's project history, particularly leadership roles in student organizations and hackathons, suggests strong initiative and collaboration skills. Their progression through various ML engineering roles at Amazon indicates adaptability and a capacity for continuous learning and problem-solving in complex environments. The detailed descriptions of impact (e.g., 'improving interaction metrics by 14%') highlight a results-oriented approach.