Yichen Lin 林奕辰

Education

Publications

1. In review

   2026

   TritonGym: A Benchmark for Agentic LLM Workflows in Triton GPU Code Generation

   Yue Guan*, Yichen Lin*, et al. * equal contribution

   Under review ICML 2026 Agentic LLMs

   Summary

   Benchmarks tool-augmented LLM workflows on Triton GPU code generation and frames the project around measurable systems behavior rather than demo-only prompting wins.

   BibTeX

   @misc{guan2026tritongym,
     title  = {TritonGym: A Benchmark for Agentic LLM Workflows in Triton GPU Code Generation},
     author = {Guan, Yue and Lin, Yichen and others},
     note   = {Under review at ICML 2026},
     year   = {2026}
   }

   Copy BibTeX

   Visit the shelf Request review
2. Unlocked 2023

   2023

   (Re)²H₂O: Autonomous Driving Scenario Generation via Reversely Regularized Hybrid Offline-and-Online Reinforcement Learning

   Haoyi Niu*, Kun Ren*, Yichen Lin, et al.

   IEEE Intelligent Vehicles Symposium (IV), 2023 Autonomous Driving

   Summary

   Explores scenario generation for autonomous driving with a hybrid offline-and-online reinforcement learning setup, focusing on robustness and controllable scenario construction.

   BibTeX

   @inproceedings{niu2023re2h2o,
     title     = {(Re)$^2$H$_2$O: Autonomous Driving Scenario Generation via Reversely Regularized Hybrid Offline-and-Online Reinforcement Learning},
     author    = {Niu, Haoyi and Ren, Kun and Lin, Yichen and others},
     booktitle = {IEEE Intelligent Vehicles Symposium (IV)},
     year      = {2023}
   }

   Copy BibTeX

   Visit the shelf

Experience

• 

  Research quest

  Samsung Semiconductor Summer research intern · San Jose

  Architecture for agentic AI workloads: runtime optimizations for agent pipelines, and simulation of how future accelerators should be designed to run them.

  Objectives (4)

  Research on agentic AI systems from the hardware side: how agent workloads behave, and what that implies for accelerator design.

  • Developed and evaluated runtime optimizations for agentic AI pipelines, and explored their hardware design implications through simulation, toward accelerators that better support agent workloads.
  • Applied agentic AI methods to build a test-debug framework prototype for MLOps.
  • Explored multi-agent coordination, structured action spaces, adaptive memory for long-term reasoning, and trust and human-AI collaboration; designed and ran experiments to evaluate the approaches.
  • Documented findings and presented insights to the research team.

  Keywords:Rewards Agentic workloads Accelerator design Runtime optimization Multi-agent systems MLOps

  Turn in quest

  Jun – Sep 2026

• 

  Research quest

  Picasso Lab, UC San Diego CSE Research intern

  Built a CXL system simulator for large-model communication; solely developed the lab websites (yufeiding.ucsd.edu, picasso-lab.com) and set up a RAG pipeline for reading papers.

  Objectives (2)

  Worked on systems support for large-model infrastructure and research tooling in an academic lab setting.

  • Developed a CXL system simulator for large model communication.
  • Sole developer of Prof. Ding's homepage and the Picasso Lab website; used RAG to parse academic papers.

  Keywords:Rewards CXL simulator RAG Lab websites

  Turn in quest

  Mar 2024 – Feb 2025

• 

  Industry quest

  Metabit Quantitative developer intern

  Sped up data parsing and added streaming reads to the internal AI platform.

  Objectives (1)

  Focused on backend-facing performance and streaming improvements for an internal AI platform.

  • Optimized data parsing and added streaming read support for AI Platform.

  Keywords:Rewards Data parsing Streaming reads

  Turn in quest

  Sep – Nov 2024

• 

  Industry quest

  Tencent, TiMi Studio Game developer intern

  Client work on the Monster Hunter mobile game, including voice-controlled teammates.

  Objectives (1)

  Built gameplay-adjacent client features and explored voice-driven teammate interaction in a production game environment.

  • Developed Monster Hunter mobile game client with voice-controlled teammates.

  Keywords:Rewards Monster Hunter mobile Voice-controlled teammates

  Talk Turn in quest

  Jun – Jul 2024

• 

  Industry quest

  Disney+ Hotstar Algorithm developer intern

  Optimised the search page and fine-tuned the recommendation model for TPUs.

  Objectives (1)

  Worked on ranking and search quality, with emphasis on recommendation and serving efficiency.

  • Optimized search page, fine-tuned recommendation model for TPUs.

  Keywords:Rewards Search page Recommendation model TPUs

  Turn in quest

  Mar – Jun 2024

• Lark

  Industry quest

  ByteDance, Lark Backend developer intern

  Built the AskAI assistant for sales-data analysis on Redis and RocketMQ.

  Objectives (1)

  Implemented backend support for AskAI workflows using message queues and caching systems in a sales-analysis context.

  • Developed AskAI assistant using Redis and RocketMQ for sales data analysis.

  Keywords:Rewards AskAI Redis RocketMQ

  Turn in quest

  Jun – Nov 2023