Ecommerce ops · Production

Mercury: eBay's Agentic AI Platform for LLM-Powered Personalized Recommendation Experiences

The problem

Scaling LLM-powered recommendation experiences to serve hundreds of millions of eBay customers across billions of listings is both challenging and costly, and LLMs alone cannot access eBay's dynamic real-time inventory.

Workflow diagram · grounded in source
1
User shopping signal received
trigger
“Whether a user is being recommended products from the latest personalized trends or receives recommendations based on their current or previous shopping missions”
2
RAG data integration
ai_action
“integrate as many relevant sources as we can, such as user and listing information as well as vast amounts of scraped publicly available content from the internet that is updated periodically via Common Crawl. Moreover, Mercury has the a…”
3
LLM query expansion
ai_action
“A pattern we developed uses a query expansion mechanism that can turn any topic into an expanded set of queries and products recommended by the LLM via its learned knowledge or its understanding of a topic via a RAG based approach”
4
Text-to-listing retrieval
ai_action
“The Listing Matching Engine's Text-to-Listing Retrieval phase dynamically matches each LLM-generated product name or search query with eBay's current listings, combining several retrieval approaches to ensure highly relevant results”
5
Anomaly detection and filtering
validation
“Anomaly Detection and Filtering to maintain relevance and accuracy. Given the large volume and diversity of eBay listings, this stage ensures quality by filtering out results that don't meet the expected explicit or inferred parameters”
6
Personalized ranking
ai_action
“the remaining listings are ranked according to a personalized model that tailors results to each user. The ranking process considers the user's previous interactions, shopping behaviors, and preferences”
7
Recommendations delivered to user
output
“the Listing Matching Engine connects LLM generated product ideas to live eBay listings, delivering a highly relevant shopping experience”
Reported outcome

Mercury enables eBay to deliver highly accurate, recent, and contextually relevant product recommendations at scale, setting a new standard for AI-driven large-scale e-commerce solutions.

Reported metrics
NRT execution latencyas little as hundreds of ms
Recommendation relevancehighly accurate, recent, and contextually relevant recommendations
Reported stack
MercuryRAGLLMsVector DatabasesKNNBERTCommon CrawlLangGraphGoogleGCPAzureOpenAI
Source
https://www.linkedin.com/pulse/mercury-agentic-ai-platform-llm-powered-experiences-ebay-chowdhury-kka8e/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Mercury enables eBay to deliver highly accurate, recent, and contextually relevant product recommendations at scale, setting a new standard for AI-driven large-scale e-commerce solutions.

What tools did this team use?

Mercury, RAG, LLMs, Vector Databases, KNN, BERT, Common Crawl, LangGraph, Google, GCP.

What results were reported?

NRT execution latency: as little as hundreds of ms; Recommendation relevance: highly accurate, recent, and contextually relevant recommendations (source-reported, not independently verified).

How is this ecommerce ops AI workflow structured?

User shopping signal received → RAG data integration → LLM query expansion → Text-to-listing retrieval → Anomaly detection and filtering → Personalized ranking → Recommendations delivered to user.

Related ecommerce ops cases