Search and AI-assisted retrieval

Improving search speed by 99.67% while increasing result relevance

Modernized search for an enterprise RFP platform using LLM-based search and retrieval strategies, improving search speed by 99.67% while significantly increasing the relevance of returned results.

Outcome

99.67% improvement in search speed

The completed work improved search speed by 99.67% while significantly increasing the relevance of returned results. The improvement came from applying LLM-based strategies to the search and retrieval process rather than simply adding an AI interface to the existing search system.

Situation

The system and the situation

Search was an important part of an enterprise RFP workflow. The existing approach needed to return useful information more efficiently while improving the relevance of the results presented to users.

Constraints

What the work had to respect

  • The search capability existed inside an established production platform.
  • Improvements had to fit the existing application and data environment.
  • Search quality and response performance both mattered.
  • LLM-based functionality could not compensate for weak retrieval or inefficient system design.
  • Changes needed to be understandable and supportable by the engineering team.

Technical work

How the work moved forward

  • Analyzed the existing search and retrieval workflow and improved the search architecture and data flow.
  • Applied LLM-based search strategies to retrieval and result selection within the established platform.
  • Contributed to backend performance improvements and production integration while balancing relevance with search speed.
  • Supported measurement and validation of the result without exposing confidential implementation details.

Result

What changed

The completed work improved search speed by 99.67% while significantly increasing the relevance of returned results. The improvement came from applying LLM-based strategies to the search and retrieval process rather than simply adding an AI interface to the existing search system.

Privacy choices