Project: GETTR Search Engine Redesign
Role: Product lead, system architect
Problem: Users couldn’t reliably find each other due to rigid, literal search logic based on ElasticSearch. The system required near-exact matches for usernames or handles, lacked typo tolerance, and ignored social context.
Solution:
Unified and normalized usernames and handles (case-insensitive, transliterated, punctuation-free)
Added prefix and fuzzy matching to handle typos and partial input
Built a social graph-based ranking model that prioritized results based on connection proximity (friends, second-degree network)
Introduced real-time activity signals and trending result caching to enhance speed and relevance
Impact:
+62% increase in successful user lookups
40% reduction in retry attempts
Autocompletion latency under 150ms
Improved onboarding and retention through more accurate, personalized discovery
In 2021, when we’re building a social platform from scratch, it's easy to overlook just how important search is—until it becomes the bottleneck to user engagement.
At GETTR, we launched fast. The early focus was on platform stability, freedom of expression, and viral loops around video and trending content. But as our user base exploded past the 1 million mark, a major usability issue surfaced:
Users couldn’t find each other.