CASE STUDY

Improving Biomedical Search Performance and Reliability

Innovent diagnosed query bottlenecks and corrected a caching defect causing out-of-memory failures in a biomedical discovery platform. The engagement also addressed relevance, memory allocation, and architecture sizing.

Client

Biomedical research discovery platform

Industry

Biomedical research

The Challenge

A biomedical research discovery platform used machine learning and search to help users explore a large collection of scientific articles. Search performance had become a concern, with some searches and facet calculations taking multiple seconds to complete.

The platform supported complex queries across article titles, abstracts, and enriched scientific concepts. Broad searches could match millions of documents, making retrieval and the calculation of facets for concepts, authors, and journals particularly demanding.

The organization needed to understand the causes of slow searches, evaluate relevance configuration, and review its architecture and node sizing. Memory failures during index updates added an operational reliability problem to the performance challenges.

Innovent’s Approach

Innovent reviewed the search environment across data ingestion, enrichment, schema design, query processing, caching, and infrastructure. The investigation distinguished the effects of query complexity, matching document counts, and cached versus uncached requests so recommendations could address specific causes.

The team identified unnecessary scoring as a significant source of query latency. Results were explicitly sorted by publication date or another value defined in the application, so relevance scores were being calculated without affecting the displayed order. Innovent tested a filtering approach that avoided those calculations while preserving the requested sort.

In controlled testing, a representative complex query that previously took approximately one second completed in less than 200 milliseconds using the filtering approach. Most simple queries tested with that approach completed in less than 100 milliseconds. These findings provided a concrete basis for query processing recommendations.

Innovent also diagnosed and corrected a cache warming defect that caused excessive memory churn during reindexing and contributed to out-of-memory failures. Memory allocation was subsequently adjusted to take advantage of the correction.

The architecture and resource-sizing review considered heap allocation, cache sizes, garbage collection, and how available resources were used. Innovent evaluated opportunities to reduce oversized allocations and recommended further testing before making broader infrastructure changes.

Facet analysis showed that latency increased sharply when calculations covered large matching document sets. Innovent evaluated options such as sampling and distributing work across smaller portions of the collection, identifying approaches for further validation rather than assuming they would deliver a particular improvement.

The engagement also included relevance engineering guidance covering field and term weighting, synonyms, special characters, stop words, and scientific language processing. Recommendations addressed representative use cases, testing strategies, and the use of interaction analytics to inform ongoing evaluation.

The Result

The confirmed correction to cache warming addressed a cause of excessive memory consumption and out-of-memory failures during reindexing. Adjusted memory allocation aligned resource use with the corrected behavior.

Query testing demonstrated that avoiding unused scoring calculations could reduce a representative complex search from approximately one second to less than 200 milliseconds. The engagement documented this opportunity alongside recommendations for facet performance, relevance configuration, and architecture sizing.

The organization gained a clearer understanding of how query behavior, caching, memory management, and data volume interacted. Innovent combined a targeted reliability correction with measured testing and architectural guidance to support informed decisions about the platform’s continued development.

Make search work better for your organization

Talk with Innovent about improving search relevance, reliability, and performance, or planning your next implementation.