Improve website search relevance by 60% and empower sales with faster, smarter results

Our first step was to index key course data fields to give the Elasticsearch engine a richer dataset for analysis and better findability. We indexed fields such as the Course Title, Subtitle, Description, Overview, Objectives, Outlines, and Job Role Mapping with the Course allowing Elasticsearch to leverage structured inputs for stronger semantic understanding and keyword matching.
 

Later, we extended this by implementing relationship mapping between courses and related entities including certifications, blogs, webinars, products, vendors, case studies, e-books, and solutions. This relational indexing strengthened the engine’s understanding of content hierarchy and improved search result categorization and relevance for users.

Once the data foundation was in place, we moved to enhance search intelligence using a hybrid semantic and keyword-based model. The initial Elasticsearch setup performed well contextually but produced overly broad results, often missing exact matches. To fix this, our Technical Manager, Abhimanyu Singh, proposed a hybrid search approach combining semantic understanding with keyword-based precision.

​

This integration balanced contextual relevance with exact keyword accuracy, leading to a more refined search experience. The first version of this hybrid model initiated our UAT phase, where we began to analyze search behavior, measure improvement, and plan further fine-tuning. This also helped me understand about what is Semantic vs keyword search. 

During UAT, we needed a way to fine-tune search relevance and control the weightage between semantic and keyword logic. Collaborating with Technical Manager, Abhimanyu Singh, I led the creation of the Boost Value Optimization Tool an internal interface that allowed our Course Management, Sales, and Marketing teams to adjust semantic and keyword weights for different course fields such as title, subtitle, description, details, outline, and job roles.
 

Teams could test various combinations (e.g., increasing keyword weight for “course title” or reducing semantic influence for “description”) to bring relevant results to the top. We collected data for 60+ high-priority keywords, averaged the effective boost values per field, and applied them to Version 1 of the search engine. This iteration produced noticeably better ranking accuracy and relevance consistency, validating our hybrid tuning model. Learning how to implement Boost value optimization in Elasticsearch was exciting form me. 

Alongside backend improvements, I focused on enhancing the usability of the search tool. My UX analysis revealed major gaps lack of filtering options, absence of auto-suggestions, no voice search capability, and limited responsiveness across devices. Moreover, weak mapping between categories like courses, certifications, and blogs often led to “No Results Found” errors for over 80% of keywords.
 

To resolve this, I redesigned the search interface and flow, introducing robust filters, auto-suggestions, and voice search, while improving mobile responsiveness and overall visual hierarchy. The improved design reduced search errors, enhanced discoverability, and created a more seamless, delightful user experience.

After implementing the optimized search system, I conducted live demonstrations with key stakeholders across Course Management, Sales, Marketing, and Technology teams. The goal was to validate improvements, showcase results, and encourage active participation in further refinement.
 

Stakeholders were invited to test keywords, report any relevancy issues, and share feedback directly through a continuous improvement channel. This collaborative feedback loop ensured the search tool evolved dynamically maintaining high relevance, user satisfaction, and business alignment over time.