PhD Student · UC Riverside

Ruoxi Zhao

Applied Mathematics & Agentic AI researcher. Developing AI and mathematical methods for high-impact problems across science, healthcare, and finance.

Pronunciation: Rowshee Zhao

AIAgentic WorkflowsLLM BenchmarkingScientific MLIn-Context LearningSDEsDynamical SystemsInverse Problems

Ruoxi Zhao with the Scotty the Bear statue at UC Riverside — With Scotty · UC Riverside

Ruoxi Zhao presenting her research poster — Presenting my research

4th

Year PhD Student

11

Research Projects

8+

Talks & Posters

Who I am

About

I’m a Ph.D. student in Applied Mathematics at UC Riverside, working on reliable LLM agents and rigorous AI evaluation, and also interested in stochastic dynamical systems and deep learning.

I build benchmarks for coding agents, including Backtrader-Bench, which received the Short Oral Paper Award at the FinLLM Workshop at IJCAI 2026. My current work studies how the screening gate shapes question difficulty in adversarial benchmark construction. Most recently, at Voaige, I ran a controlled study of whether perturbing expert routing in Mixture-of-Experts models adds useful reasoning diversity at test time.

I teach undergraduate mathematics at UCR and received the Graduate Division’s Outstanding Teaching Assistant Award for 2025 to 2026. Outside research, I love cats, and I enjoy crocheting, hiking, and snowboarding.

Programming: Python (proficient)MATLABR
Agentic AI: Custom Agentic WorkflowsAdversarial LLM BenchmarkingPrompt DesignTool UseMCQ GenerationReal-Time Data IntegrationCross-Model Evaluation
Machine Learning: In-Context LearningTransformersBayesian InferenceCNNsComputer Vision
Libraries & Frameworks: PyTorchNumPyPandasMatplotlibYOLOv8SAM
Developer Tools: GitDockerLaTeX
Languages: EnglishMandarin Chinese (proficient)
Interests: CrochetHikingSnowboardingPainting

Background

Education

University of California, Riverside

Riverside, CA

Department profile ↗

Sep 2023 to Present

· Ph.D. in Mathematics (In Progress) · expected June 2028
· Master of Science in Applied Mathematics · awarded June 2025

University of California, Merced

Merced, CA

Aug 2019 to May 2023

· B.S. in Applied Mathematical Sciences, Computational & Data Science · GPA 3.80/4.0
· Minor in Management Business and Economics (MBE)

★ Outstanding Graduating Student in Applied Mathematics · 2023

Scholarship

Publications

Backtrader-Bench: Benchmarking LLM Agents on Algorithmic Trading with Self-Generated MCQs

Published

Ruoxi Zhao, Maziar Raissi

★ Short Oral Paper Award · FinLLM 2026

FinLLM Workshop @ IJCAI 2026, Bremen, Germany · presented Aug 2026 · arXiv:2608.11232View on arXiv ↗Workshop ↗

Award certificate for Backtrader-Bench: Benchmarking LLM Agents on Algorithmic Trading with Self-Generated MCQs

Intuitive Surgical SurgToolLoc and SurgVU Challenges Results: 2022 to 2025

A. Zia, M. Berniker, … R. Zhao, … et al. (SurgToolLoc / SurgVU Challenge consortium)

★ 3rd Place · SurgToolLoc 2023 (Team ANL-Surg)

MICCAI Endoscopic Vision Challenge, Vancouver, Canada · arXiv:2305.07152View on arXiv ↗

More manuscripts in preparation.

Recognition

Honors & Awards

★

Outstanding Teaching Assistant Award · Graduate Division (TADP)

University of California, Riverside

2025 to 2026

Certificate for Outstanding Teaching Assistant Award · Graduate Division (TADP)

★

Outstanding Graduating Student in Applied Mathematics

University of California, Merced

2023

★

Chancellor's Honor List

University of California, Merced

2020 to 2023

★

Frances M. Benton Scholarship

University of California, Merced

2020 to 2023

★

Edward & Marion Doherty Fund for Yosemite Internship Program

University of California, Merced

2020 to 2022

★

School of Natural Sciences Dean's Honor List

University of California, Merced

2019 to 2023

★

Shriram Scholarship · Stockton Scholars

University of California, Merced

2019 to 2023

Portfolio

Research Experience

Apr 2025 to Present

Graduate Research Assistant · Agentic AI

University of California, Riverside · with Dr. Maziar Raissi · Riverside, CA

Paper: Backtrader-Bench (arXiv) ↗FinLLM @ IJCAI 2026 workshop ↗

Active

·Lead an adversarial benchmark-construction pipeline for LLM coding agents, in which a generator model writes multiple-choice questions about the backtrader library together with a program that computes each answer, so every label is verified by execution.
·Designed tool and no-tool screening gates that keep a question only if models fail it without tools and solve it by running code, and study how the screening rule shapes question difficulty.
·Evaluate frontier models with and without tools in network-blocked sandboxes, one attempt per question.
·Built a multi-backend agent runner across the Claude, Codex, Cursor, Copilot and Antigravity CLIs, with resumable campaigns, spend limits and per-stage cost accounting.

Agentic AIBenchmarkingLLMs

Presented at

FinLLM Workshop @ IJCAI 2026Bremen, Germany · Aug 2026Video Talk
Southern California Applied Mathematics Symposium (SOCAMS)Loyola Marymount University, Los Angeles, CA · Apr 2026Poster

Poster

Apr 2026 to Present

Graduate Researcher · Water-Treatment Modeling

UC Riverside · collaboration with Western University (Dr. Domenico Santoro) · Riverside, CA

Active

·Built a Gaussian-process / ODE hybrid model for water-treatment disinfection, producing calibrated posterior contact-time (CT) estimates while preserving EPA-compliant CT semantics.
·Derived an exact closed-form (Tricomi U) solution for microbial survival under Gamma-distributed contact time and kill rate, reducing the model to three interpretable parameters.
·Proved via Jensen's inequality that deterministic CT calculations are systematically optimistic, linking both tools into a workflow that converts plant sensor data into survival probabilities.

Gaussian ProcessesInverse ProblemsOptimization

Jun 2026 to Sep 2026

Research Intern

Voaige · Sunnyvale, CA (Hybrid)

Complete

·Studied test-time router perturbation in Mixture-of-Experts reasoning models as a source of solution diversity, as an alternative to raising sampling temperature, on competition-level math.
·Ran controlled comparisons over noise scale, target layer and noise subspace, matching each arm to stock routing by expert flip rate.
·Analyzed the router input space with PCA and SVD across several open-weight MoE model families.
·Built a vLLM routing plugin and a multi-machine experiment queue running on up to 39 A100 GPUs.

Mixture-of-ExpertsLLM ReasoningLinear Algebra

Jan 2026 to Oct 2026

Graduate Researcher · In-Context Operator Networks

University of California, Riverside · with Dr. Siting Liu · Riverside, CA

Complete

·Adapted and fine-tuned In-Context Operator Networks (ICON), a transformer-based scientific ML framework that solves 19+ ODE/PDE types zero-shot from in-context demonstrations, applied neural operator learning to Alzheimer's Disease neuroimaging and biomarker data from ADNI.
·Engineered end-to-end training pipelines on HPC clusters using JAX/CUDA, generating synthetic ODE datasets as TFRecords and training multi-shot ICON models evaluated at 1, 3, and 5 in-context demonstrations.
·Studied the mathematical foundations of transformer attention mechanisms and kernel methods.

In-Context LearningTransformersScientific ML

Jun 2024 to Sep 2024

Research Assistant

Lawrence Berkeley National Laboratory, Mathematics Dept. · Berkeley, CA

Matched via Sustainable Horizons Institute (SRP) ↗

Complete

·Analyzed and preprocessed heterogeneous X-Ray Diffraction (XRD) datasets to enable quantitative materials optimization.
·Developed objective functions and evaluated synthesis outcomes using statistical metrics.
·Automated data processing workflows using Python and Pandas.

Materials ScienceOptimization

Presented at

Southern California Applied Mathematics Symposium (SOCAMS)UC Riverside, Riverside, CA · Apr 2025Poster
Berkeley Lab WD&E Summer SymposiumBerkeley, CA · Aug 2024Poster

Poster

Jun 2023 to Sep 2023

Research Assistant

Argonne National Laboratory · with Dr. Neil Getty · Lemont, IL

Paper: SurgToolLoc and SurgVU Challenges Results 2022 to 2025 (arXiv) ↗MICCAI SurgToolLoc 2023 Challenge ↗Matched via Sustainable Horizons Institute (SRP) ↗

Published

·Surgical tool localization and detection in robotic surgery based on endoscopic video frames.
·Utilized Segment Anything Model (SAM) with text and box prompts; trained YOLOv8.
·Team ANL-Surg finished 3rd place in 2023 MICCAI SurgToolLoc challenge.
·Presented research poster at Argonne Lab's Learning on the Lawn event.

Computer VisionRobotics

Presented at

SIAM Conference on Mathematics of Data Science (MDS24)Atlanta, GA · Sep 2024Poster
14th World Congress of Robotic Surgery (CRSA)Chicago, IL · Sep 2023Talk · Poster
Argonne Learning on the LawnLemont, IL · Sep 2023Poster

Poster

Sep 2022 to May 2023

Research Assistant · Dental Cavity Detection

Learning Aligned Employment Program, UC Merced · Merced, CA

Complete

·Predicted dental cavities using color photos and Convolutional Neural Networks.
·Built data frames by converting photos to matrices with Pandas and NumPy.
·Applied Gaussian blur to experiment with its effect on prediction accuracy.

Computer VisionHealthcare

Presented at

UC Merced Research SymposiumMerced, CA · 2023Poster

Poster

Sep 2022 to May 2023

Research Assistant · Transcription Factor Binding Kinetics

CREST Fellowship, UC Merced · Merced, CA

Complete

·Quantified how transcription factor binding kinetics depend on sequence.
·Applied statistical methods and K-mer analysis across multiple datasets.

BioinformaticsStatistics

Presented at

CREST SymposiumUC Merced, Merced, CA · 2023Talk

Slides

Jun 2022 to Aug 2022

Research Internship

Lawrence Berkeley National Laboratory, Joint Genome Institute · Berkeley, CA

Complete

·Integrated single-cell and spatial datasets using Tangram and PyTorch.
·Predicted cell type locations in spatial voxels; cell segmentation via Cellpose.
·Presented at Berkeley Lab's final symposium.

GenomicsML

Presented at

SIAM Conference on Mathematics of Data Science (MDS22)San Diego, CA · Sep 2022Poster
Berkeley Lab JGI Intern Symposium & JGI User MeetingJoint Genome Institute, Berkeley, CA · 2022Talk · Poster

Poster

Slides

May 2022 to Jun 2022

Data Science Challenge

Lawrence Livermore National Laboratory · Livermore, CA

Complete

·Built binary classifiers using molecular descriptors and 3D CNNs to find COVID-19 antivirals.
·Processed 2D/3D datasets (HDF5) with Pandas, NumPy, and PyTorch.

BioinformaticsDeep Learning

Presented at

LLNL Data Science Challenge SymposiumUC Merced, Merced, CA · Jun 2022Talk

Slides

May 2021 to Jun 2021

Data Science Challenge

Lawrence Livermore National Laboratory · Livermore, CA

Complete

·Worked on planetary defense: classified stars vs. galaxies and captured asteroids from Palomar Observatory data.
·Applied CNNs with PyTorch; presented results to LLNL scientists and engineers.

AstrophysicsComputer Vision

Presented at

LLNL Data Science Challenge SymposiumUC Merced, Merced, CA · Jun 2021Talk

Slides

Experience

Professional

Apr 2025 to Present

Treasurer

SIAM Student Chapter, UC Riverside

·Manage the chapter's budget and finances, coordinating funding for student activities and conference travel.
·Represented the chapter as a student representative at the 2026 SIAM Annual Meeting (AN26).

LeadershipSIAMService

Sep 2023 to Present

Teaching Assistant

Mathematics Department, UC Riverside

·Assisted instruction for undergraduate courses: precalculus, calculus, discrete mathematics, differential equations, and linear algebra.
·Led discussion sections and office hours; graded assignments and exams.

TeachingMathematicsMentoring

Sep 2023 to Jan 2024

Data Science Consultant

Graduate Quantitative Methods Center (GradQuant), UC Riverside

·Led workshops on data science fundamentals for UCR graduate students and postdoctoral scholars.
·Provided one-on-one coding and quantitative analysis support via remote consultations.

Data SciencePythonConsulting

Sep 2022 to Oct 2022

Data Analysis Intern

Businessolver & Open Avenues

·Built an interactive visualization tool in Python (Pandas, Matplotlib, NumPy) to analyze user-chatbot interactions.
·Compared inbound/outbound user trends and performed word frequency analysis on tweet data.

Data AnalysisVisualizationNLP

Jun 2021 to Aug 2021

Shriram Summer Fellow

Reinvent Stockton Foundation & Stockton Scholars

·Created landscape analysis of Summer Bridge programs for incoming 9th graders across California.
·Presented findings to the Executive Board via Google Slides.

ResearchPolicyEducation

Mar 2021 to Apr 2021

Project with Purpose

Merced County Arts Council

·Conducted case study research on the development of Arts & Culture Districts in demographics similar to Merced and their civic benefits.
·Provided recommendations to the MCAC on strategies for designating Downtown Merced as an Arts & Culture District.
·Created a PowerPoint and video presentation to introduce the concept and process to interested stakeholders.

ResearchCommunityPolicy

Talks & Posters

Presentations & Conferences

Aug 2026

FinLLM Workshop @ IJCAI 2026

Bremen, Germany

“Backtrader-Bench: Benchmarking LLM Agents on Algorithmic Trading with Self-Generated MCQs”

See research ↓

Jul 2026

SIAM Annual Meeting (AN26)

Student Volunteer

Cleveland, OH

Attending as a student representative for the UCR SIAM Student Chapter.

May 2026

ACM Conference on AI and Agentic Systems (CAIS)

Attending

San Jose, CA

Apr 2026

Southern California Applied Mathematics Symposium (SOCAMS)

Loyola Marymount University, Los Angeles, CA

“Optimization and Evaluation of Autonomous AI Agents in Sequential Decision Environments”

See research ↓

Apr 2025

Southern California Applied Mathematics Symposium (SOCAMS)

UC Riverside, Riverside, CA

“Optimization Formulations for Self-Driving Inorganic Materials Design”

See research ↓

Sep 2024

SIAM Conference on Mathematics of Data Science (MDS24)

Atlanta, GA

“AI Foundation Models for Surgical Tool Localization”

See research ↓

Aug 2024

Berkeley Lab WD&E Summer Symposium

Berkeley, CA

“Optimization Formulations for Self-Driving Inorganic Materials Design”

See research ↓

Sep 2023

14th World Congress of Robotic Surgery (CRSA)

Chicago, IL

“AI Foundation Models for Surgical Tool Localization”

See research ↓

Sep 2023

Argonne Learning on the Lawn

Lemont, IL

“Surgical Instrument Localization using Language and Vision Foundation Models”

See research ↓

2023

CREST Symposium

UC Merced, Merced, CA

“Understanding Transcription Factor Binding Kinetics Using Statistical Learning Methods”

See research ↓

2023

UC Merced Research Symposium

Merced, CA

“Machine Learning Applications for Dental Cavity Prediction”

See research ↓

Sep 2022

SIAM Conference on Mathematics of Data Science (MDS22)

San Diego, CA

“Integrating Single Cell Data with Spatial Transcriptomics using Machine Learning”

See research ↓

2022

Berkeley Lab JGI Intern Symposium & JGI User Meeting

Joint Genome Institute, Berkeley, CA

“Integrating Single-Cell Data with Spatial Transcriptomics Using Machine Learning”

See research ↓

Jun 2022

LLNL Data Science Challenge Symposium

UC Merced, Merced, CA

“Binary Classification of Test Compounds with Molecular Descriptors and 3D Atomic Representations”

See research ↓

Jun 2021

LLNL Data Science Challenge Symposium

UC Merced, Merced, CA

“Detecting Asteroids using Image Classification”

See research ↓

Mar 2020

Women in Data Science (WiDS)

Attending

Livermore, CA

Off the clock

Beyond Research

When I'm not doing research, you'll find me hiking, snowboarding, crocheting, or hanging out with my cats, Matt and Ricky.

Ruoxi Zhao at Sequoia National Park — Sequoia National Park

Ruoxi Zhao snowboarding — Snowboarding

Ruoxi Zhao holding a funnel cake at a night festival — Funnel cake

Ruoxi Zhao bouldering at a climbing gym — Climbing

Matt the cat — Matt

Ricky the cat — Ricky

Crochet work by Ruoxi Zhao — Crochet

Crochet work by Ruoxi Zhao — Crochet

Crochet work by Ruoxi Zhao — Crochet

Get in touch

Contact