I’m Lily Zhang, a Tech Lead and Research Scientist in the Bay Area, California. My research focus: agentic post-training and AI training data, including RL environments. I’ve published at CVPR, ICCV, NeurIPS, RAL [1] [2] [3] [4] [5], and gave keynotes at NVIDIA GTC and IEEE IROS.

I like turning research into products people use, open-sourced: SofaGenius (Anthropic hackathon finalist, top 15/13k), Frontend Slides (25K+ GitHub stars), and the Deep-Seek research agent (500+ GitHub stars). At Latitude AI, I build LLMs, multimodal LLMs, and AI agents for physical AI (Modal GTC panel on agentic post-training).

Recent work: C-Guard, constitution-grid data-efficient RL alignment (COLM 2026 ER); AURA, RLVR reward shaping from a hierarchical failure taxonomy; Eureka: feature engineering as agentic code generation w Alibaba Cloud, SFT + RL post-trained AI-infra agent deployed in production (NeurIPS-W 2025, DASFAA 2026). Publishing as Xianling Zhang.

AI Research

Find all published research here.

A Constitution-Grid Instrument for Data-Efficient RL Alignment (C-Guard)

X Zhang · COLM 2026 ER

Post-trained safety guard model with RL, automatic synthetic data generation for safety alignment

Paper Blogpost Code Dataset

Eureka: Feature Engineering as Agentic Code Generation

H Li, R Jia, X Wu, Y Qian, Z Zheng, X Zhang · NeurIPS-W 2025, Proceedings of the 31st International Conference on Database Systems for Advanced Applications (DASFAA 2026, Oral)

SFT + RL post-trained AI-infra agent deployed at Alibaba Cloud: +16% demand, 91% adoption.

Paper Poster Project Page

AURA: Automatic Reward Shaping from a Hierarchical Failure Taxonomy

X Zhang

Build verifiable failure taxonomy for RL reward shaping in Multi-turns harbor terminal environment.

Blogpost Code (Soon) Data (Soon)

SuperGeneral: Compositional Tool Environments for Long-Horizon Agents

X Zhang

Learn meta skills. Evaluates four frontier models on tool use, composition, and creation on law, consulting, and investment banking domain datasets.

Project Page Code

SIMBAR: Single Image-Based Scene Relighting for Effective Data Augmentation for Automated Driving Vision Tasks

X Zhang, N Tseng, A Syed, R Bhasin, N Jaipuria · Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022

Project Page Paper

Deflating Dataset Bias Using Synthetic Data Augmentation

N Jaipuria, X Zhang, R Bhasin, M Arafa, P Chakravarty, S Shrivastava, S Manglani, VN Murali · Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2020

Paper

Advisory Boards

Industry Advisory Board, Texas A&M University

Robotics / MXET program.

Board Meeting

Industry Advisory Board, University of Delaware

Supporting young researchers and shaping strategic directions for the CAR Robotics Lab.

Guest Editor, ACM Transactions on Internet of Things

Autonomous Driving special issue.

Call for Papers