AI Agent & LLM Developer (Data Scientist)

Haoyang Yu

La Jolla, CA • 909-***-**** • **********@*****.***

EDUCATION

Harvard University, Cambridge, MA Sep 2026 - Expected Dec 2027 Master of Science in Data Science

University of California, San Diego, La Jolla, CA Sep 2022 - Mar 2026 Bachelor of Science in Data Science Minor in Cognitive Science & Mathematics GPA: 3.87/4.00 Achievement: Alice C. Tyler Scholarship (24-25); UCSD TOWN & GOWN (24-25, 25-26); Provost’s Honors for every quarter. Relevant Courses: Systems for Scalable Analytics, Data Visualization & Management, Probability & Statistics TECHNICAL SKILLS

Programming: Python, JavaScript, MATLAB, SQL

Machine Learning: PyTorch, scikit-learn, LLMs, Deep Learning, Transformer, Fine-tuning, Post-training, RLHF, SFT LLM / Agent: OpenClaw, Codex, Prompt Engineering, Context Engineering, Agent Workflow EXPERIENCE

PANGO VALVE LLC Remote - Houston, TX July 2026 – Present AI Application Assistant

● Developed an enterprise AI Agent platform for expense management using ERPNext, Frappe, Python, and LLMs, supporting expense applications, reimbursements, payment requests, multi-step approvals, and intelligent workflow automation.

● Designed intelligent workflow orchestration, role-based permission systems, and financial knowledge models supporting AI-assisted approval decisions for 40–80 enterprise users.

● Built LLM-powered document understanding pipelines integrating OCR, invoice parsing, document completeness verification, duplicate detection, policy compliance reasoning, and approval recommendation generation.

● Prototyped conversational AI interfaces and Agent-based task execution pipelines for enterprise expense workflows, enabling natural language interaction and future integration with external Agent frameworks. Hao AI Lab La Jolla, CA May 2024 – June 2026

Research Assistant Supervised by: Dr. Hao Zhang

● Built scalable evaluation pipelines for LLM Agents across 20+ foundation models and 10+ interactive environments, supporting 10,000+ GPU evaluation runs.

● Developed an open benchmarking platform, LMGame Bench, for evaluating LLM reasoning, decision-making, and multi-turn agent performance on Hugging Face.

● Designed modular multi-agent evaluation harnesses supporting reproducible interaction, tool execution, and multi-turn reasoning experiments.

● Co-developed Video ScienceBench, a benchmark for evaluating scientific understanding in video generation models, covering causal reasoning, temporal consistency, and physics-based scenarios. PARAMETRICS AI Remote — New York Aug 2025 – Nov 2025 Data Analyst

● Built ML pipelines for a healthcare data client, integrating clinical trial records, patient behavior logs, and genomic datasets to enable predictive modeling across 50K+ samples.

● Developed survival and logistic regression models to forecast treatment responses and adverse events, improving baseline accuracy by ~15% and supporting data-driven clinical decisions. Scripps Research - The Zorrilla Laboratory La Jolla, CA July 2024 – Dec 2024, June 2025 – Sep 2025 Research Intern Supervised by: Dr. Eric Zorrilla

● Extracted and standardized 354,000+ alcohol-related EHR records from the All of Us database by mapping ICD-9/10 codes to SNOMED CT via OMOP, aligned with the lab’s curated set of ~100 diagnoses; built SQL pipelines to merge datasets.

● Analyzed >200 behavioral assay datasets from murine stress models and hippocampal/amygdalar proteomics, applying ANOVA and regression to quantify stress-induced behavioral variance and molecular shifts.

● Designed reproducible workflows integrating data cleaning, table joins, and temporal aggregation, reducing missing/noisy entries by 30% and improving the reliability of downstream predictive models. PUBLICATIONS

Hu, L.; Huo, M.; Zhang, Y.; Yu, H.; Xing, E. P.; Stoica, I.; Rosing, T.; Jin, H.; Zhang, H.

“lmgame Bench: How Good Are LLMs at Playing Games?” arXiv preprint arXiv:2505.15146, May 2025. Accepted at ICLR 2026.

Zhang, Y.; Yu, H.; Hu, L.; Jin, H.; Zhang, H. “General Modular Harness for LLM Agents in Multi Turn Gaming Environments.” arXiv preprint arXiv:2507.11633, July 2025. Accepted at ICML 2025 Multi Agent Systems (MAS) Workshop. Yu, H.; Zhang, Y.; Hu, L.; Jin, H.; Zhang, H.

“VideoScience-Bench: Benchmarking Scientific Understanding and Reasoning for Video Generation.” arXiv preprint arXiv:2512.02942, Dec 2025. Accepted at ECCV 2026.

Contact this candidate