Shang Hong Sim

Post training research to make open models capable and reliable.

I’m a Machine Learning Engineer at Oumi, where I work on post-training for the enterprise platform. I own on-policy distillation end to end and actively work on training science for our SFT, RL, and distillation workflows.

My research asks how to make LLMs reliable as they become the way people consume information. With Trust-Align (ICLR ‘25 Oral), we built a metric for groundedness in RAG systems and an alignment dataset that teaches models to cite faithfully, and to refuse when they should. Ground-GRPO followed up with RL on verifiable rewards to cut unsupported citations.

Before Oumi, I did my M.Eng (Research) at the DeCLaRe Lab at SUTD with Prof. Soujanya Poria, writing a thesis on reliable RAG systems.

My path into tech began in the sciences. I was premed and wanted to be a doctor, and biology remains a real love of mine. That passion carried me through research in cancer bioinformatics (co-author in Nucleic Acids Research), brain-signal decoding, and protein structures in drug-resistant bacteria. The habits I built there, careful experiment design and thorough analysis, still shape how I do ML research today.

Away from the keyboard, I love baking artisanal bread and am a sucker for trying out new knitting designs. I’m also a huge nature girl, and I feel super fortunate to live in Seattle with the trails of the Pacific Northwest at my doorstep for hiking and biking. You can see what I’m up to off hours, and the full record of what I’ve done lives on my CV.