Hybrid Search Powered Search Experience for ScreenSteps

ScreenSteps is a Knowledge Operations platform that helps organizations become change resilient instead of change resistant, and protect supervisor bandwidth. They specialize in working with companies that deal with complexity and change in their procedures. Their customers include financial institutions, healthcare orgs, contact centers, law offices, back-office operations, and other teams that rely on accurate, up-to-date procedures to do their work.

As their customers' organizations grew, so did their knowledge bases-often spanning thousands of articles. Ensuring employees could find accurate, up-to-date information quickly became a mission-critical priority.

ScreenSteps users could retrieve articles easily-but the search capability wasn't nearly robust enough. They realized that they needed to handle searches better than by just using keywords. They needed to be able to find answers to questions and handle semantic connection rather than just relying on a basic keyword-based search approach.

ScreenSteps partnered with BigData Boutique, an AWS Advanced Partner and global expert in search and data infrastructure, to reimagine their search experience from the ground up.

Architecture Overview

The teams jointly designed a hybrid search architecture built on Elastic Cloud, AWS Bedrock, Nova and Cohere V4. This architecture combined multiple layers of intelligence:

• Search classification - They began classifying user queries using Nova as a router. This gave the system the ability to handle old keyword searches the "old way" and use hybrid search for questions and more complex searches.
• Hybrid Retrieval - Leveraging embeddings and semantic vector search to surface contextually relevant content.
• Question Generation - Question generation was used to improve embedding accuracy and make sure that the right documents are found for each question.

Under the hood, Question Generation using the AWS Nova model on Bedrock helped handling different types of documents using question generation and classify search terms while Elasticsearch handled hybrid search.

Key Technical Highlights

• Indexing and Data Modeling: Articles were converted into questions that served as semantic pointers to the original documents, each embedded via Cohere embeddings and indexed into Elasticsearch for high-performance hybrid retrieval.
• Inference Pipeline: A microservice architecture running within AWS Lambda and ECS orchestrated handled processing of new or modified documents.
• Evaluation and Feedback: Relevance benchmarks were continuously measured through MAP (Mean Average Precision), ensuring precision remained high even as data scaled.

“BigData Boutique’s understanding of hybrid retrieval and their ability to merge semantic and keyword search were key. The new architecture doesn’t just improve results-it changes how users trust the search itself.” - ScreenSteps Executive Team

The collaboration resulted in a measurable improvement in both system precision and user satisfaction:

• More Precise Search - Queries that previously returned dozens of results now yield only 1–3 highly relevant matches, or none when no valid answer exists.
• Smarter Knowledge Gaps - The system now highlights missing knowledge, enabling content teams to prioritize updates and fill gaps faster.
• Improved Feedback Loops - Organizations now have visibility into what users can’t find, driving more strategic content creation.

“Instead of uncertainty about whether an answer exists, teams can now take action to fill knowledge gaps-improving both the completeness and usability of their knowledge base.” - ScreenSteps Executive Team

ScreenSteps selected BigData Boutique for their deep technical expertise in Elasticsearch, semantic search and AI-powered query understanding. Their experience in building production-grade hybrid search pipelines within the AWS ecosystem made them the natural choice for this project.

“We needed more than a search vendor-we needed a partner who could understand how our users think. BigData Boutique’s mix of AI innovation and infrastructure know-how delivered exactly that.”

Following the successful rollout, ScreenSteps plans to extend this hybrid search framework to improve knowledge discoverability analytics, and user intent insights. By expanding their integration with AWS-native services, they aim to continuously refine the precision and transparency of their knowledge search experience.

“Our partnership with BigData Boutique has redefined how we think about search. It’s not just about finding answers-it’s about helping our users trust when there isn’t one.”