Shanghai Jiao Tong University
Department of Artificial Intelligence
Sep 2023 – Jun 2027
GPA: 4.0 · Cohort rank: 2/22 · Shanghai, China
Selected for the Bolè Plan (Elite Scientific Talent Program), one of 30 students among 18,584 SJTU undergraduates.
Curriculum vitae
Academic CV
Yanjie Huang 黄䶮杰
I am an undergraduate researcher at Shanghai Jiao Tong University. I build self-improving AI systems for automated research through algorithm design, with a focus on scientific synthetic data, post-training, test-time search, and learning from computational and experimental feedback.
- Education
- SJTU · Artificial Intelligence
- Expected
- June 2027
- GPA
- 4.0
- Focus
- Self-improving AI
01
Education
02
Research Experience
Stanford University
Research Assistant
Advisors: Prof. Yejin Choi, Prof. Jure Leskovec, and Dr. Fang Wu.
- Recursive Self-Improvement for ML Research and Engineering · Ongoing. Implementing and empirically evaluating recursive self-improvement approaches for kernel optimization and ML algorithm development.
- UniEvo-VL · May – Sep 2026 · Co-first author. Built the initial distributed OPSD pipeline, integrating on-policy sampling, critique-conditioned teacher feedback, LoRA updates, and evaluation; implemented ablations over noise, prompts, seeds, denoising steps, and teacher dynamics. The framework improved Qwen-Image-2512 from 0.747 to 0.808 on GenEval and from 32.97 to 35.53 on GenEval2 Soft-TIFA.
- Proteo-R2 · May 2026 – Present. Built pipelines to generate, filter, and standardize reasoning traces for text-molecule alignment, mid-training, and joint training; trained residue- and atom-level reasoning models for controllable molecular design.
- ProteoCraft · Mar – May 2026 · Co-first author. Built reproducible benchmarks across search strategies and diffusion- and hallucination-based baselines for binder and TCR design. Contributed architecture, template and domain redesign, and Monte Carlo Tree Search; reduced memory usage by 40%.
Columbia University
Research Team Leader
Advisor: Prof. Erran Li.
- Proteo-R1 · Mar – Apr 2026. Synthesized reasoning data for alignment, mid-training, and joint training, improving indexing ability by 40% and translating model understanding into controllable antibody design. ICML (Accepted).
Tsinghua University
Research Team Leader
Advisor: Prof. Yi Lin.
- PhaseFlow · May 2025 – Jul 2026. Built data and quality-control pipelines for a high-throughput phase-separation dataset and a unified model combining sequence generation with flow matching for phase-diagram prediction. Outperformed DDPM and Transformer-only baselines; implemented inference-time scaling, improving self-consistency by 30%.
Shanghai Jiao Tong University
Research Team Leader
Advisor: Prof. Shuangjia Zheng.
- EVA · Jul 2025 – Mar 2026. Developed MoE architectures with MegaBlocks and the first pretraining and mid-training code, achieving 30% lower perplexity than the dense baseline; co-curated and released OpenRNA v1 across more than 15 RNA types. Used sparse autoencoders to identify structure/function features (Spearman > 0.9) and steer generation. The best designed aptamers improved fluorescence and binding affinity by 2x over the positive control.
- Non-canonical amino acid design · May 2025 – Jul 2026. Used experimental data for DPO-based few-shot learning, improving enrichment of high-fitness proteins by 30%; built BoltzGen, Boltz2, and AlphaFold generator-predictor pipelines, yielding 2 successful designs and 14x activity improvement.
- RNA evolution · Jul 2024 – Mar 2025. Combined inverse-folding and language-model approaches for RNA design and fitness benchmarking, achieving 2.6x in vivo fluorescence improvement for multiple variants.
Massachusetts Institute of Technology
Research Assistant
Advisor: Dr. Jack Xiaoyu Chen.
- SynthArchitect · Mar – May 2025. Built an agent integrating 20+ tools, including Golden Gate and DNA Cauldron, for automated plasmid engineering and laboratory workflows.
03
Selected Research
Multimodal self-improvement · May – Sep 2026
UniEvo-VL
Co-first author. Built distributed critique-conditioned on-policy self-distillation for unified multimodal models. ICLR 2027 (Under Review).
View project →Automated ML research · Ongoing
Recursive Self-Improvement
Implementing and empirically evaluating self-improvement approaches for kernel optimization and ML algorithm development.
View project →Lab-in-the-loop learning
EVA
Team leader and first author. Connected MoE model development, mechanistic interpretability, and experimentally validated RNA design.
View project →Multimodal modeling · Inference-time scaling
PhaseFlow
Team leader and co-first author. Built sequence-and-flow modeling and inference-time scaling for controllable phase-separating protein design.
View project →Scientific reasoning synthetic data
Proteo-R1 / Proteo-R2
Built synthetic reasoning data and trained multimodal models for controllable molecular design at residue and atom resolution.
View project →Inference-time search · Efficient evaluation
ProteoCraft
Co-first author. Built search benchmarks and contributed architecture and Monte Carlo Tree Search, reducing memory usage by 40%.
View project →04
Selected Publications
UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement
Fang Wu*, Da Xing*, Yanjie Huang*, Junxi Wang*, …, Jure Leskovec†, Yejin Choi† · ICLR 2027 (Under Review)
EVA: A Long-Context Generative Foundation Model Deciphers RNA Design Principles
Yanjie Huang*, Guang Lv*, …, Yi Yang†, Shuangjia Zheng† · Nature Machine Intelligence (Accepted, Spotlight, Top 10%) · 2026
PhaseFlow: A Unified Model for Multi-scale Biomolecular Condensates Understanding and Design
Yi Zhu*, Yanjie Huang*, …, Shuangjia Zheng†, Yi Lin† · Nature (Under Review) · 2026
Energy-Adaptive Elastic Equivariant Transformer-SSM for Noise-Robust Protein Structure Representation
Zhongyue Zhang, Runze Ma, Yanjie Huang, Shuangjia Zheng† · NeurIPS 2026 (Accepted)
ProteoCraft: Proposal-Verification Search for Protein Design with Diffusion Structure Predictors
F. Wu*, H. Lin*, B. Duan*, Yanjie Huang*, Y. Li*, …, Jure Leskovec†, Yejin Choi† · ICLR 2027 (Under Review)
Capturing Natural Evolution in RNA Design for in Vivo Functionality via Genomic Foundational Models
Yanjie Huang*, B. Zhang*, …, Yi Yang†, Shuangjia Zheng† · Communications Biology (Under Review) · 2025
What Matters in RL for Diffusion Models? The Dominant Role of Noise
Da Xing, Haokai Zhao, Yanchao Li, Hanqun Cao, Yanjie Huang, …, Yejin Choi†, Fang Wu† · ICLR 2027 (Under Review)
05
Honors
- 2026National Scholarship for Undergraduate Students
Sep 2026.
- SJTUBolè Plan
Elite Scientific Talent Program, Shanghai Jiao Tong University.
- 2022Bronze Medal
Chinese Mathematical Olympiad · Dec 2022.
- 2021First Prize · International Round
International Mathematical Modeling Challenge · Jul 2021.
06
Technical Focus
Models & Training
MoE Transformers; unified multimodal and diffusion models (Qwen-Image, BAGEL); multimodal alignment; pretraining, SFT, DPO, on-policy self-distillation (OPSD), and LoRA.
Data & Evaluation
Scientific reasoning synthetic data; large-scale data processing (>150 TB); streaming quality control; benchmarking and ablation studies.
Systems
Distributed training; Megatron-LM; MegaBlocks; vLLM; agentic systems (LangChain, Qwen-Agent).
Programming
Python, C++, Shell.
PhD & research opportunities
Let's build AI that improves how research is done.
I am happy to discuss research collaborations, PhD opportunities, and research engineering roles in self-improving AI and automated research.
huangyanjie@sjtu.edu.cn
