Jinyan Su
Research Intern at Microsoft
California, United States
Summary
🤩
Rockstar🎓
Top SchoolJinyan Su is a research-focused machine learning engineer specializing in safety and alignment, with nine years of experience building models that are helpful, honest, and capable of abstention when appropriate. Currently a Research Intern at Microsoft after recent research internships at Meta and Adobe, Jinyan develops post-training interventions (SFT, RL) and prompt-steering methods to improve model robustness without degrading core capabilities like math and QA. They design synthetic-data generation, reward functions, and RL environments, and have experimented with using LLMs as judges for safety evaluation and tool-enabled reasoning. A PhD candidate at Cornell and visiting scholar at Stanford, Jinyan blends rigorous academic training with hands-on product-focused research in multi-turn dialogue and retrieval-augmented agents. Based in California, they bring a pragmatic approach to alignment problems—often tackling subtle failure modes such as stale or misleading queries by engineering targeted data and evaluation pipelines.
9 years of coding experience
Bachelor of Science - BS, Mathematics-Information and computational sciences, Bachelor of Science - BS, Mathematics-Information and computational sciences at University of Electronic Science and Technology of ChinaDoctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at Cornell UniversityVisiting Scholar, Computer Science, Visiting Scholar, Computer Science at Stanford University
English, Chinese, Spanish
Github Skills (117)
rdma10
glm10
openai-api10
lsm-tree10
multimodal10
ai10
transformer10
gpt10
rust10
database10
net-library10
llama10
storage10
reinforcement-learning10
amd10
Programming languages (9)
Github contributions (5)
pegainfer-project/pegainfer
Feb 2026 - Jul 2026
Pure Rust + CUDA LLM inference engine — no PyTorch, OpenAI-compatible, serves Qwen3 to Kimi-K2
Contributions:16 reviews, 38 PRs, 367 pushes in 5 months
cudallm-inferenceopenaipytorchrust
novitalabs/pegaflow
Jan 2026 - Jul 2026
High-performance KV cache storage for LLM inference — GPU offloading, SSD caching, and cross-node sharing via RDMA. Works with vLLM and SGLang.
Contributions:3 releases, 27 reviews, 65 PRs in 6 months
cache-storagecachinggpullm-inferencerdma
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.