HELLO, I'M
M.S. and B.S. in Computer Science from Harbin Institute of Technology, with five years of R&D experience across foundation models, search, and recommendation. My work focuses on large-scale pre-training, model scaling, long-context modeling, and efficient inference.
WHAT I DO
End-to-end experience spanning data development, model architecture, distributed training, performance optimization, and production deployment. Experienced in leading teams of 10+ engineers and collaborating on training and inference across clusters with tens of thousands of accelerators.
FEATURED WORK
Taking ideas from research prototypes to full-scale production, with measurable gains in model quality, computational efficiency, and business outcomes.
LONG CONTEXT · EFFICIENT ATTENTION
Designed a Transformer/HSTU-based long-context model for long-range dependencies, interest exploration, and long-tail behavior in recommendation sequences. Introduced variable-length Flash Attention and dynamic batching to reduce padding overhead.
PRE-TRAINING · SCALING · LARGE-SCALE TRAINING
Led a seven-person team in building the full data, modeling, training, evaluation, and deployment pipeline. Scaled dense models to 4B parameters, sparse models to over 100 trillion parameters, and training context to 110K tokens.
LLM POST-TRAINING · USER UNDERSTANDING
Built an end-to-end pipeline from user signals through LLM-based interest reasoning and label generation to downstream retrieval. Applied chain-of-thought prompting, few-shot learning, and Qwen3-based knowledge distillation to improve new and low-activity user experiences.
EXPERIENCE
Multiple performance ratings in the top 30%
Consistently top 10% · Three consecutive promotions · Department Excellence Award
EDUCATION & INTERNSHIPS
Computer Science and Technology · NLP and Human-Computer Dialogue
GPA-equivalent score: 90.75 · Top 0.5%Computer Science and Technology
GPA-equivalent score: 88.98 · Top 4%Recommendation Algorithm Engineer Intern · Jul–Sep 2020
Raised OCR copy acceptance from 90% to 95% and online exposure by 7%Algorithm Engineer Intern · Apr–Jul 2020
Improved slot-filling accuracy from 45% to 86%ACHIEVEMENTS
National Second Prize in the China Undergraduate Mathematical Contest in Modeling; ranked 18/163 and 20/93 in the CCF Language and Intelligence Technology Competition, and 16/256 in the JD Dialogue Challenge.
Over RMB 200K in scholarships; two national, one provincial, and more than twenty university-level awards.
First author of two top-tier conference papers and two journal articles; six patents, including three as first inventor; one software copyright.
LET'S CONNECT
Foundation models, search and recommendation, long-context modeling, and efficient inference.