CASE STUDY

Replacing a Strained Search Platform for Job Discovery

Innovent replaced a strained job search platform with a redundant search architecture, faster indexing, and tuned relevance. Average search response times fell below 250 milliseconds, while index preparation dropped from three hours to under 20 minutes.

Client

Recruitment marketing and talent acquisition software provider

Industry

Human Resources Technology (SaaS)

The Challenge

Searches typically took 2–6 seconds. The company needed search experts to replace the existing implementation with one that could process job data efficiently, improve matching and ranking, and maintain availability as demand grew. Its developers also needed to understand and extend the solution for additional applications.

Job information arrived through customer-provided XML feeds and was loaded into Microsoft SQL Server. The existing search process broke job content into individual terms and built database records connecting those terms to jobs and source fields. Generating and searching this data consumed substantial database resources, sometimes leaving the website unable to return results during processing.

Searches typically took 2–6 seconds. The company needed a replacement that could process job data efficiently, improve matching and ranking, and maintain availability as demand grew. Its developers also needed to understand and extend the solution for additional applications.

Innovent’s Approach

Innovent designed and implemented a replacement that moved indexing and search processing out of the production database. The team developed a search schema balancing the fields needed for matching and ranking with the information returned for website display.

A lightweight ingestion framework extracted job records from SQL Server, transformed the content, and loaded it into the search index. Parallel processing made use of the ingestion server’s capacity, while streamlined database queries reduced the burden on production resources.

Innovent tuned relevance using field based weighting and representative search scenarios. Testing addressed synonyms and linguistic behavior, as well as content processing issues involving character encoding, HTML removal, and fields containing multiple values.

The team also performed load testing against estimated peak traffic for the following two years, evaluating the implementation against anticipated growth.

The search architecture separated indexing from query serving workloads. A dedicated indexing server supplied two replicated search servers, allowing a load balancer to distribute requests from the website. This provided redundancy without requiring the production database to perform the search workload.

Innovent completed a week of hands-on knowledge transfer covering configuration, code, and alternative implementation scenarios. This prepared the client’s developers to maintain the system and extend its capabilities into other applications.

The Result

Average search response times fell from 2–6 seconds to under 250 milliseconds. The time required to prepare a searchable index dropped from three hours to under 20 minutes, eliminating the database intensive search processing that had competed with production workloads.

Full index replication to the search servers completed in less than 60 seconds. The redundant deployment separated indexing from website search traffic, while relevance tuning improved the handling of job content and varied search requirements.

Innovent delivered the project on time and within budget. The client gained a faster job search platform, a more efficient indexing process, and the knowledge needed to maintain and extend the implementation.

Make search work better for your organization

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