Back office ops · Production
Expedia Group ML Platform builds a centralized Embedding Store Service for vector similarity search
The problem
ML teams at Expedia Group faced significant engineering and integration overhead when building vector embedding use cases, with no centralized solution for storing, managing, or discovering embeddings across the organization.
Workflow diagram · grounded in source
1
Collection creation with metadata
trigger
“The metadata defined can include the associated service (the system or application that generates and/or consumes the embeddings) and the specific model used to produce them”
2
Embedding data ingestion
integration
“There are three methods available for loading data, depending on the volume and generation process”
3
Online and offline storage
output
“the service ensures that all embeddings are stored simultaneously in both the online and offline storage systems, providing robust access for various use cases”
4
Vector similarity or hybrid search
ai_action
“similarity searches can be performed to find embeddings that are most similar to a given query vector”
Reported outcome
The Embedding Store Service provides centralized vector embedding management with reduced development time, standardized APIs, and support for batch, real-time, and on-the-fly embedding workflows across Expedia Group.
Reported metrics
Development timeReduced development time and acceleration of development and iteration
Reported stack
FeastSpark
Source
https://medium.com/expedia-group-tech/powering-vector-embedding-capabilities-12e8e1480f43
Read source ↗Frequently asked questions
What did this team achieve with this AI workflow?
The Embedding Store Service provides centralized vector embedding management with reduced development time, standardized APIs, and support for batch, real-time, and on-the-fly embedding workflows across Expedia Group.
What tools did this team use?
Feast, Spark.
What results were reported?
Development time: Reduced development time and acceleration of development and iteration (source-reported, not independently verified).
How is this back office ops AI workflow structured?
Collection creation with metadata → Embedding data ingestion → Online and offline storage → Vector similarity or hybrid search.