// 01 About
Who I Am
I'm an OR/data scientist with a passion for building high-quality AI-powered optimization software. I specialize in crafting performant, scalable AI-powered optimization systems from algorithm to code.
When I'm not designing and building features, you'll find me contributing to open source, exploring the latest research bridging the gap between OR and AI/ML, or experimenting with creative coding and generative art.
// 02 Experience
Where I've Worked
My professional journey — roles, responsibilities, and impact.
Senior Operations Research Scientist
Coupa Software Inc.Major Achievements
- Integrating a LLM-powered AI agent to Coupa's supply chain prescriptions software, which allows users to run different prescriptions types in a modularized fashion based on natural language queries.
- Led the redesign and improvement of the logic of volume consolidation — a prescriptions type of Coupa's supply chain prescriptions software.
Operations Research Scientist
Coupa Software Inc.Major Achievements
- Designed and productionized a risk modeling feature in Coupa's supply chain network optimization software, powered by a mixed integer programming (MIP) model.
- Trained a ML model to automatically terminate the Gurobi solver early for network optimization runs to save compute costs without degrading solution quality.
Staff Data Scientist
Walmart Inc.Major Achievements
- Led the reformulation of a non-linear MIP-based associate workforce planning model to a linear program (LP), reducing solver runtime from hours to minutes.
- Built a reinforcement learning (RL) algorithm to dynamically balance overstaffing and understaffing costs in real-time workforce planning.
Data Scientist
Coupa Software Inc.Major Achievements
- Maintained and improved Coupa's demand forecasting software; added a seasonality detection module and a probabilistic forecast distribution fitting module.
- Led the design of the integration workflow between demand modeling and supply chain network optimization software, enabling MIP-based network design using probabilistic demand forecasts instead of deterministic historical records.
Scientist II
PROS Holdings Inc.Major Achievements
- Enhanced an MIP-based supply chain network optimization model for a dairy manufacturer by incorporating a price uncertainty risk measure into the objective function.
- Designed a personalized dynamic pricing algorithm for airline ancillary products using contextual multi-armed bandit reinforcement learning, improving customer revenue by 8%.
- Built an ensemble model for airline ticket demand forecasting on 1TB+ of passenger search history, reducing forecast error by 26% vs. an existing Bayesian model.
Research Assistant
Virginia TechMajor Achievements
- Proposed a heteroscedastic t-process metamodel as an alternative to Gaussian process for simulation optimization, applied to periodic review inventory control.
- Developed load forecasting algorithms leveraging spatial and temporal data with k-means clustering, deep learning, and Gaussian process regression; designed optimal online sampling strategies via reinforcement learning.
- Proposed a probabilistic online trajectory prediction algorithm for unmanned aerial vehicles, supported by the National Science Foundation.
// 03 Skills
What I Work With
A snapshot of my technical toolkit — bars reflect relative comfort, not rank.
Optimization & Operations Research
Machine Learning & Data Science
Tools & Platforms
// 04 Projects
What I've Built
A selection of side projects — personal and open source.
ORPilot
A LLM-powered AI agent for operations research (OR) modeling that translates natural-language problem descriptions into mathematical formulations and executable optimization models. This Bridges the gap between domain experts and solver implementation — no OR modeling expertise required.
RL-Based Inventory Optimization
A reinforcement learning framework for periodic-review inventory control under stochastic demand. Trained a policy network to dynamically adapt order quantities, outperforming classical (s, S) policies on a range of demand distributions.
Transformer-Aided Optimization
A transformer model trained for learning solutions to MIPs for faster solving. Applied transformer's attention mechanisms to combinatorial optimization, learning solution heuristics end-to-end without hand-crafted features. Tested on facility location problem instances.
// 05 Activities
Professional Activities
Conferences, academic service, and writing.
Conferences
- Seminar hosted by SAP Germany"Reinforcement Learning for Inventory Optimization"Remote · 2023Presenter
- INFORMS Annual Meeting" A Markowitz-Based Model for Optimizing Supply Chain Networks with Risks Related to Price Uncertainty"Anaheim, CA · 2021Session Chair
- INFORMS Annual Meeting"An Adaptive Streaming Data Sampling Strategy via Reinforcement Learning for Efficient Load Forecasting"Seattle, WA · 2019Presenter
- INFORMS Annual Meeting"An Adaptive Data Sampling Scheme for Bandwidth Limited Residential Hourly Load Forecasting"Phoenix, AZ · 2018Session Chair
- The 49th North American Power Symposium"An Adaptive Communication Scheme for Bandwidth Limited Residential Load Forecasting"Morgantown, WV · 2017Presenter
Reviewer Service
- Applied Sciences
- Mathematics
- Energies
- Sustainability
- IEEE Access
Blog
- How to build an Agentic AI Database Assistant for Supply Chain Systems
- Can ChatGPT solve knapsack problems?Towards Data Science
- Transformer Aided Supply Chain Network DesignTowards Data Science
- Reinforcement Learning for Inventory Optimization Series III: Sim-to-Real Transfer for the RL ModelTowards Data Science
- Reinforcement Learning for Inventory Optimization Series II: An RL Model for A Multi-Echelon NetworkTowards Data Science
- Reinforcement Learning for Inventory Optimization Series I: An RL Model for Single RetailersTowards Data Science
- Some Thoughts on Synergies between Operations Research and Machine LearningTowards Data Science
// 06 Contact
Get In Touch
Open to interesting conversations, collaborations, and opportunities.
The best way to reach me:
guanx92@vt.edu© 2025 Guangrui Xie · Built with Next.js + Framer Motion
Designed & developed with ♥ using vibe coding