Applied AI · Retail · Supply Chain · Decision Intelligence
I’m Harshavardhan Nannur, a Staff Data Scientist at Walmart. Across 13+ years in retail and supply chain, I’ve built production systems that help teams understand complexity, evaluate alternatives, and move from prediction to action.
Selected impact
Production systems built around difficult operating decisions. Each story begins with the problem, then explains the contribution and the outcome.
An AI-assisted root-cause and workflow system that helps planners understand supplier performance and move from investigation to action.
Risk models that focus inspection capacity on the distribution-center days and purchase-order lines most likely to need attention.
Mixed-integer optimization systems that translate operating constraints into practical allocation, logistics, and workforce decisions.
A discrete-event simulation that makes complex container-flow and capacity trade-offs visible before operational decisions are made.
A human-in-the-loop claims concept combining guided claim creation, image analysis, risk scoring, and conversational supplier-quality insights, targeting a reduction in resolution time from 96 hours to 24 hours.
Multimodal GenAI · Fresh claimsProfessional perspective
My work sits at that transition: connecting models to operating context, constraints, workflows, and human judgment.
A working principle
Durable value comes from connecting data, prediction, context, alternatives, action, and feedback. That principle links my work across machine learning, optimization, simulation, and agentic systems.
Patents & innovation
Original work presented with precise status, scope, and a clear explanation of why the contribution matters.
US20250103976A1
US20260017606A1
A Python and SimPy discrete-event simulation approach for evaluating capacity, renovation, disruption, and demurrage and detention scenarios in import distribution-center operations.
Ideas
Short, substantive field notes on the design choices that determine whether enterprise AI becomes useful.
Enterprise AI creates value only when a prediction is connected to context, constraints, alternatives, and a responsible path to action.
Read the field note ↗Useful enterprise agents must understand the workflow, assemble evidence, respect decision boundaries, and know when human judgment is required.
Read the field note ↗The hardest work begins after model selection: data quality, adoption, latency, feedback, workflow design, and trust determine whether AI changes an operation.
Read the field note ↗
Speaking
A practical discussion of discrete-event simulation for capacity planning, operational scenarios, and cost-risk analysis in import distribution-center networks.
Selected recognition
Recognition conferred upon approximately the top 1% of Walmart corporate employees.
Enterprise-wide competition recognizing applied generative-AI innovation.
Recognized for the AI-Assisted Claims multimodal generative-AI solution.
The professional through-line
My career has moved from customer analytics and forecasting into optimization, simulation, production machine learning, and agentic AI. The constant is a focus on what the organization needs to decide—and how technology can make that decision better.
Read the full storyAvailable for relevant invitations
I welcome focused opportunities to share, evaluate, or advance serious technical work with the broader professional community.
Conferences, universities, panels, podcasts, and practitioner forums.
↗02Applied AI, machine learning, operations, and technology innovation.
↗03Enterprise AI, retail technology, supply chains, and agentic systems.
↗04Collaboration where production experience complements technical research or strategy.
↗Start a conversation
Let’s exchange ideas about retail AI, decision intelligence, agentic systems, optimization, or simulation.