JZ.

Jingyuan Zhu.

I like building agentic AI systems and the evals that keep them honest πŸ€–πŸ©ΊπŸ“ˆ

gh in @

2026 -
LI
I'm a machine learning engineer at LinkedIn on the Agentic Intelligence Platform (CoreAI), working on evaluation for InSeek β€” LinkedIn's agentic search over the Economic Graph, serving 1B+ members. I focus on modeling and evaluation methods for agentic systems, including recursive self-improvement of agents across product surfaces.
2026 -
PM
In parallel, I'm a founder engineer (MTS) at PediaMed.AI Research Lab β€” pediatrics foundation models, child-centered agentic AI, and cognitive LLM/VLMs inspired by how children learn. I build the research systems: evaluation infrastructure, data pipelines, and the lab's public platform.
2024 - 2026
UP
M.S.E. in Data Science at the University of Pennsylvania. Along the way I interned at Amazon (summer 2025), where I built an end-to-end LLM annotation assistant on Bedrock for a financial-analysis platform β€” it cut manual annotation time by 60% and hit 10,000+ uses in peak weeks β€” and I built a distributed task scheduler in Go for Wharton Research Data Services that handles 10,000+ backtesting tasks/second with 99.99% delivery reliability.
2020 - 2024
WU
B.S. at Washington University in St. Louis with a dual major in computer science and mathematics (GPA 3.97). Summer internships at Hundsun Technologies β€” RAG with custom BM25 + vector retrieval, distributed LLM training with DeepSpeed/LoRA, vLLM serving at 3Γ— throughput β€” and at Isvision, where YOLOv5 models I tuned for railway inspection went from F1 0.56 to 0.90.
bio
Jingyuan Zhu is a machine learning engineer at LinkedIn (Agentic Intelligence Platform, CoreAI) and a founder engineer at PediaMed.AI Research Lab. Jingyuan's research spans vision-language models, embodied AI, and children's health, with publications at NeurIPS, CVPR, and ECCV, and degrees from the University of Pennsylvania (M.S.E., Data Science) and Washington University in St. Louis (B.S., computer science & mathematics).

projects
AutoConference is an OpenReview-style conference platform where every role β€” author, reviewer, area chair, program chair β€” is played by an AI agent, and humans only observe. The goal is to answer "what is actually broken in peer review, and what would help" with reproducible data instead of anecdotes: every state change lands in an append-only event log that exports as a public dataset. TypeScript/Next.js/Postgres, v0.1, in development.
?
Medical Doctor's Last Exam is "Humanity's Last Exam," adapted for medicine: a benchmark of questions frontier models still get wrong. Submitted questions survive a quality gate, small- and large-model gauntlets, an AI judge, and expert human review before entering the benchmark. FastAPI/React/Postgres with a durable job queue, live API-cost telemetry, and medical imaging support. Live.
A distributed task scheduler for financial backtesting at Wharton Research Data Services: a Go orchestrator handling 10,000+ tasks/second with at-least-once semantics, self-healing workers coordinated through Redis heartbeats, resource-aware scheduling across heterogeneous nodes, and gRPC APIs instrumented with Prometheus/Grafana. 99.99% delivery reliability.
pediamedai.com is the lab's public site β€” research, publications, team, and the workshop pages for ICLR 2025 and CVPR 2026 β€” which I hand-maintain as a static Hugo site on GitHub Pages.
publications
Fine-Grained Preference Optimization Improves Spatial Reasoning in VLMs
NeurIPS 2025
Y. Shen, Y. Liu, J. Zhu, X. Cao, X. Zhang, Y. He, W. Ye, J. M. Rehg, I. Lourentzou
PALM: Progress-Aware Policy Learning via Affordance Reasoning for Long-Horizon Robotic Manipulation
CVPR 2026
Y. Liu*, J. Zhu*, Y. Mo, G. Li, X. Cao, J. Jin, Y. Shen, Z. Li, T. Yu, W. Yuan, F. Ding, I. Lourentzou (* equal contribution)
Decoding Children's Gait Behavior
ECCV 2026
Y. Shen, B. Li, M. Huang, Y. Liu, X. Cao, J. Jin, Z. Li, A. Liu, J. Kim, J. Zhu, L. Fangzhou, J. Cao, J. Chen, I. Lourentzou, J. M. Rehg
A Survey on Embodied AI for Pediatrics
CVPR 2026 Workshop (CV4CHL)
Y. Shen, Q. Zhang, Y. Zhang, J. Zhou, H. Yang, Y. Wang, J. Zhu, O. K. Susladkar, Y. Tian, M. Huang, J. Chen, X. Cao
Immersive Augmented Reality Music Interaction through Spatial Scene Understanding and Hand Gesture Recognition
IEEE ICIR 2025
S. Huang, G. Shen, Y. Song, Y. Kang, J. Zhu
misc unsorted