Somesh Rahul

Title of the Talk:
From Static Pixels to Shoppable Intelligence: Building Generative Visual Commerce at Walmart

Abstract :
Text-based product search is a 30-year-old interface paradigm built for a different era of shopping. This talk presents the computational architecture behind Walmart’s production visual commerce layer: a suite of interconnected AI systems that let customers discover, visualise, and purchase products through images, context, and generative scenes rather than keywords.
Three production systems anchor the discussion. Shop the Background applies real-time object detection and inventory retrieval to transform lifestyle imagery into interactive Shoppable storefronts, enabling customers to buy what they see instantly. Explore Styles uses large-scale generative AI to synthesise aspirational room scenes at scale, placing products in contextual environments across tens of thousands of SKUs without a photographer or studio. Visual Search and Shop Similar Items close the loop by enabling image-based retrieval as a first-class discovery path.
Each system surfaces distinct computational challenges: retrieval accuracy under ambiguous visual queries, latency constraints at retail-scale inference, scene generation quality control across a high-volume content pipeline, and the cost economics of running generative models against a catalog of millions of items. The talk traces how these capabilities were sequenced into production, and the evaluation frameworks built to measure quality in the absence of ground-truth labels.
Validation across 70 AI-synthesized environments found that 9 in 10 users rated the generated scenes at the top of the usefulness scale, clearing the bar to expand the system across the catalog. Attendees will leave with a concrete architectural model for building multi-modal AI product systems at retail scale, along with the evaluation and cost frameworks required to justify them.

Bio:
Somesh Rahul is a Staff Product Manager at Walmart Global Tech, where he leads generative AI, computer vision, and spatial computing products that shape how millions of customers discover and visualize products online.
His work spans visual search, generative scene synthesis, and AR-based shopping experiences deployed at retail scale.
He is an active member of IEEE and ACM, has spoken at venues including the IEEE New Era AI Summit and CSCT 2025, and guest lectures at universities on applied AI in commerce. He holds a computer science degree and writes on the intersection of AI, spatial computing, and ambient retail.