Justin Zhao
I turn LLM research into robust agentic AI applications that work in the real world.
I am a Ph.D. Candidate with an interdisciplinary background in computer science and social science, building multi-agent orchestration workflows for research problems while studying their broader societal implications. Over the past 8 years, I have developed software across research and production, with the last 4 focused on NLP and AI infrastructure; my applications have earned 1000+ stars on GitHub and scaled to hundreds of thousands of user interactions across platforms. Across this work, my consistent principle is that robust AI systems require systematic evaluation, reproducible benchmarking, and bias-aware validation before deployment at scale. As I complete my dissertation, I am looking to own projects that are truly innovative and impactful, achieve substantial production adoption, and drive tangible commercial value.
Timeline
From a price monitor script in 2019 to production LLM systems and AI products today, each phase added engineering depth, product experience, and operational rigor.
Democratizing LLM Development: NoCode Platform
Architected a graph-based multi-agent orchestration framework, replaced 4 weeks of training with minutes of drag-and-drop on a canvas to build customized LLM workflows, and benefited 1200+ users monthly; listed as a trending space on Hugging Face. Reduced processing time by 86% through model routing and by dispatching concurrent LLM batches through Redis-backed Celery workers. Solved bugs reported by users and balanced requests against available resources to keep the product meeting real community needs. 500+ stars on GitHub. 70+ stars on Hugging Face.
Price Monitor 2.0: Agentic Trading System
Lowered the barrier to quantitative trading for non-technical users by architecting an agentic trading framework with customizable orchestration layers that let agents autonomously fetch data, build context, make and review decisions, and cooperate with each other. Coached 500+ non-technical users by demoing platform mechanics and risk analysis during high-volatility periods, presented the product to prospective investors, and translated feedback into product iterations. 200+ stars on GitHub.
Research Journey
Ph.D. Research Assistant, National Science Foundation, University of Texas at Dallas. Automated a data cleaning pipeline using LLMs to process 10GB+ of noisy corpora, reducing human annotation time by 95% and converting a 3-month manual labor into 4 hours of parallelized GPU computation on HPC. Designed an AI model evaluation framework to measure LLM biases, providing guidance for 200+ researchers to use LLMs responsibly. Coached 5+ early-career researchers on LoRA and RLHF best practices, partnered with multiple research groups to translate requirements into technical specifications, and established annotation and data collection guidelines for international collaborators.
M.S. Computer Science
University of Colorado Boulder training in machine learning and cloud computing, applied to the systems and products built above.
Teaching Assistant, University of Texas at Dallas
Designed courses and teaching materials for 5+ courses, lectured at Math and Coding Summer Camps and Research Initiatives, and taught lab sessions in Quantitative Methods, Inferential Statistics, Knowledge Mining, Information Management, Data Visualization, and Machine Learning for Text Data, directly benefiting 500+ students and researchers. Analyzed and documented students' failure patterns, evaluated course outcomes, and drafted curriculum standards for the organization.
Operations Analyst, TikTok, ByteDance
Optimized advertising strategies for stakeholders by making data-driven decisions, increasing traffic for targeted accounts by 300% on average. Built automated pipelines to record, analyze, and market accounts and campaigns, reducing 30 hours of weekly manual work to 1 hour per week. Led a cross-functional task force to manage client requests during campaigns, translating urgent needs into operational solutions, keeping deliverables on track under pressure, and synthesizing outputs into stakeholder-ready insights presented to senior leadership.
Where It Began: Price Monitor v1.0 and distributed VPN network, 2019
Built the first iteration of Price Monitor in 2019, a rule-based system for tracking prices and surfacing actionable signals. The same core problem that drove the 2019 version, making market information legible and actionable without constant manual monitoring, is what the 2026 agentic system solves at a fundamentally different level of autonomy.