Coding agents
Work on making coding agents more reliable and efficient for SDLC: better code context, efficient tool use, cleaner task delegation, and task-level failure analysis and evaluations.
Yongjun Chen
about
Recent Work on AI Systems and Products
These days, finding joy in understanding model behavior, turning capability into systems, and shaping systems into everyday products people can trust and enjoy.
Apple Intelligence
Worked on Apple Intelligence initiatives end to end. Efforts spanned benchmark training infrastructure in the early days, production data synthesis and mixture design, post-training and model distillation, and ReAct-style long-horizon reinforcement learning for agentic models. Contributed to AI product launches, including Writing Tools and Reply Suggestions.
A Bit of Earlier Research
Earlier work before the AI wave accelerated in late 2022 covered sequential recommendation and AutoML at Salesforce AI Research, with a 4th-place finish in the CIKM 2021 AutoML challenge. Even earlier exploration concentrated on deep learning methods for image pixel-wise prediction.
Blogs & Notes
Working notes and occasional two cents on research, systems, products, big tech, startups, and life.
currently working on coding agents at Augment Code; previously worked on Apple Intelligence at Apple (2022.09 - 2026.01), sequential recommendation and AutoML at Salesforce AI Research (2019.08 - 2022.09), and spent a short period at Dell EMC (2019.01 - 2019.07) on cloud data store related engineering work.
before that, received an M.S. in Computer Science (2016-2018) from Washington State University, with a thesis on advanced deep learning methods for image pixel-wise prediction, and a B.S. in Statistics (2012-2016) from Huazhong University of Science and Technology.
strive to stay focused while adaptive, collaborative while determined, with high agency and low ego.