ZI WANG • STAFF MACHINE LEARNING ENGINEER
ZI WANG
STAFF MACHINE LEARNING ENGINEER
Agentic AI Reinforcement Learning Real-Time ML Distributed Systems Optimization *************@*****.*** +1-415-***-**** linkedin.com/in/guzzii Modesto, CA, United States PROFESSIONAL SUMMARY
Staff Machine Learning Engineer with 13+ years of experience spanning production machine learning, distributed systems, mathematical optimization, and applied AI research. Progressed through five ML engineering levels at Instacart, leading business- critical initiatives in agentic AI, personalized recommendations, reinforcement learning, dynamic pricing, and real-time delivery prediction. Architected large-scale production ML systems, led engineering teams, and translated predictive models into business outcomes. Ph.D. in Industrial Engineering & Operations Research with patented innovations and peer-reviewed research in distributed optimization.
TECHNICAL SKILLS
AI & Machine Learning: LLMs, Multi-Agent Systems, Agent Orchestration, LLM Evaluation, Recommendation Systems, Ranking, DNN, CNN, RNN, Behavioral Sequence Modeling, Probabilistic Forecasting Reinforcement Learning & Optimization: Conservative Q-Learning (CQL), Offline RL, Counterfactual Policy Evaluation, Doubly Robust Estimation, Mixed-Integer Programming, Dynamic Pricing ML Infrastructure & MLOps: Griffin, Maple, Ray, Ray AIR, KubeRay, Airflow, Spark, Hadoop, Distributed Training, Online Inference, Monitoring, CI/CD
Programming & Frameworks: Python, R, SQL, TensorFlow, Keras, PyTorch, scikit-learn, NumPy, SciPy, PyArrow Cloud & Infrastructure: AWS, S3, EKS, Kubernetes, Docker, Redis, ScyllaDB, Jenkins, Git Decision Science: Contribution Profit Optimization, Conversion Modeling, Time-Series Forecasting, A/B Testing, Model Calibration, Gurobi, CPLEX, GAMS
PROFESSIONAL EXPERIENCE
INSTACART · San Francisco Bay Area · Remote
Staff Machine Learning Engineer, Agentic Marketing & Personalization Jun 2025 – Oct 2026 Project: Storefront Pro — Customer Campaigns
Architected and deployed an end-to-end multi-agent LLM marketing platform, replacing weeks of manual campaign curation with near-instantaneous generation; orchestrated targeting, content, and evaluation agents for personalized, inventory-aware campaigns.
Designed an isolated dual-loop architecture with independent LLM critics and deterministic guardrails for pricing, inventory, and stale context; reduced invalid campaign payloads to approximately 0.5% before CRM execution.
Integrated Maple distributed LLM infrastructure using PyArrow, S3, asynchronous batches, retries, and rate-limit handling; leveraged published platform baselines of 2.6 tasks/second and large-batch completion under 12 hours.
Technically led customer behavior sequence modeling using temporal embeddings of purchase patterns, replenishment, baskets, and incentives; reduced prediction log-loss by 12% versus the legacy baseline.
Developed offline reinforcement learning with Conservative Q-Learning (CQL) to jointly optimize promotional timing and discount magnitude against customer lifetime value, subsidy costs, and margins.
Implemented counterfactual policy evaluation with doubly robust and importance-sampling estimators, demonstrating 17% estimated offline policy-value improvement versus legacy rules. Staff Machine Learning Engineer, Service & Pricing Apr 2024 – May 2025 Project: Real-Time Delivery ETA & Contribution Profit Optimization
Led multi-year optimization of delivery tiers, ETA windows, and checkout options, balancing conversion, delivery-fee revenue, fulfillment cost, and expected contribution profit.
Architected online self-correcting ETA calibration using streaming fulfillment actuals and localized residuals to adjust predictions without offline retraining.
Developed P10–P90 probabilistic ETA bounds and approximately 1.2 volatility buffers to adapt commitments during demand peaks, shopper shortages, and traffic disruptions.
Implemented risk-aware delivery eligibility guardrails targeting 90% on-time delivery, suppressing unreliable options above an 8% predicted lateness threshold.
ZI WANG • STAFF MACHINE LEARNING ENGINEER
Introduced zero-minute late tolerance and 20-minute early-arrival windows for SLA measurement, reliability monitoring, and operational cost controls.
Integrated calibrated ETA predictions into contribution-profit-aware checkout decisions considering Priority surcharges, conversion, incremental fulfillment costs, and margins. Senior Machine Learning Engineer II Aug 2022 – Mar 2024 · Remote
Advanced distributed ML training and inference for ETA forecasting, order fulfillment, and marketplace optimization using Python, Ray, parallel processing, and feature engineering.
Modernized production ML workflows with Griffin 2.0, Ray AIR, KubeRay, AWS EKS, Kubernetes, Airflow, and Docker for scalable training, orchestration, deployment, and monitoring.
Developed marketplace supply-demand forecasting using time-series models, statistical optimization, online inference, and experimentation.
Senior Machine Learning Engineer I Aug 2020 – Jul 2022 · Remote
Proposed and led the architecture and implementation of a real-time delivery ETA service with 3 ML engineers, replacing fixed- duration promises with dynamic per-order predictions.
Deployed low-latency inference APIs with real-time order-level features, achieving 99.9% production serving uptime.
Established prediction reliability and monitoring workflows for changing marketplace conditions. Machine Learning Engineer II Nov 2018 – Jul 2020 · Remote
Developed item-availability classification models using historical shopping data and feature engineering to estimate store-level product availability.
Contributed to item-substitution recommendation and ranking through product similarity, behavioral features, and offline model evaluation.
Supported fulfillment capacity and demand forecasting with time-series modeling and large-scale operational data analysis. MONSANTO COMPANY · Greater St. Louis Area · Remote Data Scientist Apr 2016 – Nov 2018
Project: Predictive Breeding & Genomic Analytics
Led 4 data scientists developing production predictive and prescriptive ML for plant breeding, guiding architecture, optimization, and engineering quality.
Productionized DNN, CNN, RNN, Random Forest, and Gradient Boosting models in Python, R, TensorFlow, Keras, and PyTorch on genomic, environmental, and trial data.
Contributed to ML-driven corn breeding improvements associated with company-wide outcomes of 5 more varieties evaluated and one-year faster product development.
Built large-scale genomic processing, feature, and training pipelines with Python, SQL, Spark, and Hadoop to enable reproducible high-dimensional analysis.
Championed Git, testing, Jenkins CI/CD, Docker, documentation, and reproducible pipelines; presented validated results to senior R&D leadership.
PENN STATE UNIVERSITY · State College–DuBois Area
Research Assistant — Distributed Machine Learning Aug 2013 – Sep 2016
Co-developed Distributed First-Order Augmented Lagrangian (DFAL) algorithms for decentralized large-scale ML optimization across 10 nodes.
Benchmarked asynchronous optimization, proximal gradient descent, and randomized block coordinate descent, showing up to 99% lower execution times than ADMM in selected experiments.
Evaluated graph-coordinated distributed ML on 2,000-dimensional datasets, improving scalability and reducing synchronization overhead.
Co-authored ICML 2015 research on distributed first-order optimization, with benchmark execution times as low as 5 seconds in selected configurations.
Consultant at Supply Chain Research Center · University Park, PA Feb 2013 – Aug 2013
Consulted with Ingersoll Rand on outbound logistics, designing mixed-integer programming models for LTL transportation, inventory holding, and delivery-time tradeoffs.
Evaluated pooled distribution and milk-run strategies in GAMS, identifying a network design with 30% projected expense reduction versus baseline.
ZI WANG • STAFF MACHINE LEARNING ENGINEER
MONSANTO COMPANY · Greater St. Louis Area
Operations Research Scientist Intern May 2015 – Aug 2015 Project: Predictive Genomic Selection & Smart Breeding Optimization
Developed deep learning models for genetic sequence data using Python, R, NumPy, and SciPy to predict outcomes and prioritize high-value plant crosses.
Co-developed smart breeding pipelines with scikit-learn, SQL, Gurobi, and CPLEX, combining genomic predictions with constrained decision optimization.
Contributed operations research and predictive analytics associated with later patent family US11728010B2. EDUCATION
Penn State University Aug 2011 – Jun 2016
Ph.D., Industrial Engineering & Operations Research Stevens Institute of Technology Sep 2008 – May 2010 Master’s Degree, Nanotechnology
Beijing Institute of Technology Sep 2004 – Jun 2008 B.S., Materials Science