From Vespa.ai to MongoDB Atlas: Modernizing Enterprise Search Without Touching the User Experience
A Fortune 10 healthcare and insurance group migrated a member-facing care cost search platform off Vespa.ai onto MongoDB Atlas — consolidating document, vector, and search workloads on a single managed platform, retiring scarce specialist skills, and clearing a path for enterprise-wide decommissioning.
$ cat case-study/details.yaml
The Problem.
The client operates a member-facing care cost search experience that lets users look up a condition or procedure and see a personalized care path — the necessary treatment steps and the average cost of each, calibrated to the member's specific healthcare plan.
The product was running on Vespa.ai. As a search engine, it was capable; as enterprise infrastructure inside a Fortune 10 healthcare group, it had accumulated cost the business no longer wanted to carry:
1. Specialist skills sparsely available. Search/SRE expertise to operate Vespa.ai at this scale was expensive, scarce, and concentrated in a small number of engineers.
2. Operational stack fragmented. Separate search infrastructure ran alongside document storage and ML ranking pipelines, with schema synchronization, custom indexing operations, and cross-system consistency to maintain.
3. Tech debt blocking AI cloud adoption. The legacy footprint slowed the broader move to managed, cloud-native services across the enterprise.
The migration goal was straightforward to state and difficult to deliver: move the product onto MongoDB Atlas with full functional and performance parity, transition all core search, autocomplete, and ML ranking capabilities into a production-grade environment, with zero impact on the user experience at cutover — and in doing so, set up Vespa.ai for enterprise-wide decommissioning beyond this first product itself.
What we built.
Quest1 designed and delivered the migration as a side-by-side validation rollout — both stacks running in parallel through a phased cutover, with continuous comparison of search relevance and performance until the new stack was production-ready. Five engineering moves anchored the work:
The Impact.
The migration delivered the operational simplification the business was after. The product now runs on a single managed platform — Atlas absorbing the document store, vector search, and ranking footprint that had previously required three integrated systems and the engineering skill base to maintain them. The team eliminated dedicated search infrastructure, custom indexing operations, and schema synchronization pipelines from the operational stack.
- A consolidated, managed-services search and ranking stack on MongoDB Atlas, replacing the multi-system Vespa.ai footprint.
- Atlas Vector Search powering the embedding-based ML ranking for care path retrieval.
- A side-by-side validation pattern — and the observability built around it — that the client's platform team can apply to future search and retrieval migrations.
- A path forward for enterprise-wide Vespa.ai decommissioning beyond the first product, with this engagement as the first production proof.
- Reduced dependency on specialist search/SRE skill sets that were expensive and scarce inside the enterprise.