Shengzhi (Alex) Li

Shengzhi (Alex) Li

I am a Research Engineer at Meta, working on post-training for the Llama model family. Previously, I was at Reka AI (contributing to Reka-flash-3), TIFIN (JPMorgan-backed fintech), and Cala Health (3 patents on neurostimulation devices, now Medicare-covered). My research focuses on Large Language Models, Multi-modal LLMs, and Alignment.

Email Scholar GitHub LinkedIn HuggingFace

News

[Mar 2026]Released IntelliAsk: learning to ask high-quality research questions via RLVR. Outperforms GPT-5 and Gemini 2.5 Pro.
[Jun 2025]Joined Meta as Research Engineer on the Llama post-training team.
[2025]Selected as ICLR 2025 Top Reviewer.
[2025]Reviewing for ICML 2025, AISTATS 2025, IEEE SSCI 2025.
[2024]Selected as NeurIPS 2024 Top Reviewer (top 10%).
[Aug 2024]Paper on multi-modal preference alignment accepted at ACL 2024 (Main Conference, top 10% score).
[Aug 2024]Joined Reka AI as Member of Technical Staff; contributed to Reka-flash-3.
[Aug 2024]Invited talk at 27th Mars Society Convention, University of Washington.
[Apr 2024]Promoted to Staff LLM Data Scientist at TIFIN.
[Aug 2023]Released SciGraphQA dataset (downloaded 6K+ times).
[2023]Co-organized Rare Disease AI Hackathon at Stanford School of Medicine.
[2022]Three patents on wearable neurostimulation systems granted.

Publications

IntelliAsk: Learning to Ask High-Quality Research Questions via RLVR
Karun Sharma, Vidushee Vats, Shengzhi Li, Yuxiang Wang, Zhongtian Sun, Prayag Tiwari
arXiv 2026
Trains a reward model on 15.5K+ ICLR review questions and uses RLVR to generate expert-quality peer review questions, outperforming GPT-5 and Gemini 2.5 Pro.
Paper Project
Multi-modal Preference Alignment Remedies Regression of Visual Instruction Tuning on Language Model
Shengzhi Li, Rongyu Lin, Shichao Pei
ACL 2024 (Main Conference, Long Paper)
Identified and addressed modality degradation in MLLMs, achieving superior results with only 5K training examples vs 80K+ in prior work.
[Paper] [Code]
Abstract2Appendix: Academic Reviews Enhance LLM Long-Context Capabilities
Shengzhi Li, Rongyu Lin, Kittipat Kampa, Bohang Li, Shichao Pei, Bo Han
Under review at COLING 2025
[Paper]
Wavelet Guided Visual State Space Model and Patch Resampling Enhanced U-shaped Structure for Skin Lesion Segmentation
Shuangxi Feng, Xianglin Chen, Shengzhi Li
IEEE Access, 2024
SciGraphQA: A Large-Scale Synthetic Multi-Turn Question-Answering Dataset for Scientific Graphs
Shengzhi Li, Nima Tajbakhsh
arXiv 2023 (Downloaded 6K+ times, cited 52 times)
Created the largest open-source scientific graph VQA dataset (295K samples, 13x larger than ChartVQA).
[Paper] [Dataset] [Code]

Experience

Meta Research Engineer Jun 2025 – Present
Reka AI Member of Technical Staff Aug 2024 – Jun 2025
TIFIN Staff LLM Data Scientist / Senior NLP Scientist Aug 2023 – Aug 2024
Cala Health Data Scientist Oct 2019 – Aug 2023
Zeesan Biotech Data Scientist Jun 2018 – Oct 2019

Education

WestCliff University M.S. Computer Science, Summa Cum Laude 2021 – 2022
Stanford University Graduate Certificate in AI 2021
UC Berkeley B.A. Applied Mathematics & B.A. Physics 2013 – 2018

Patents

Academic Service

Top Reviewer: NeurIPS 2024 (top 10%), ICLR 2025

Conference Reviewer: NeurIPS, ICLR, ICML, AISTATS, IEEE SSCI, TMLR

Journal Reviewer: IEEE Trans. Intelligent Vehicles (8.2), Neural Networks (7.8), CVIU (7.8), Applied Soft Computing (7.2), Computers & Electrical Engineering (4.3)

Invited Talks