A subscription-based crafting-commerce platform in the US
Custom semantic search that lifted conversion 30% and serves 100M+ monthly queries at sub-50ms latency.
+30%
Higher conversion rate
40%
More relevant results found
<50ms
Search latency at 100M+ queries/mo
The subscription-based platform required manual tagging of each image, taking several minutes per asset and a dedicated team. This capped content scalability and created ongoing operational inefficiencies.
Manual tagging offered no guarantee of consistency, leading to poor search experiences. Users struggled to find relevant content for their projects, adding friction and reducing subscription renewals.
Users needed to search highly specific crafting terminology, copyright content, character names, and nuanced project requirements that generic search engines could not understand.
With more than 100M queries a month and a fast-growing content library, the existing infrastructure could not scale while keeping search quality high and response times under 50ms.
We trained specialized deep-learning models on millions of in-house crafting assets to capture nuanced terminology and domain knowledge. Built with Python and PyTorch, the models understand crafting contexts, copyright content, and character names that generic models miss.
We implemented OpenSearch-based vector indexing and similarity queries tuned for rapid recall, letting users find relevant assets 40% more often through semantic understanding rather than keyword matching.
We deployed on AWS SageMaker with Inferentia chips for low-latency inference at scale, serving hundreds of requests per second at sub-50ms latency and replacing the legacy search engine with no parallel system required.
We embedded an A/B testing framework for continuous UX improvement, validating search-algorithm changes in real time and keeping the experience and conversion rates improving.
The platform achieved 30% higher conversion rates through a dramatically improved search experience. Users find the right assets 40% more often, driving engagement and renewals while the system handles 100M+ queries a month at consistent sub-50ms response times.
Automating content understanding eliminated manual tagging bottlenecks. The solution scales with content growth while maintaining quality, reducing operational costs and freeing specialized teams for higher-value work.
Contact us
We build production AI systems end to end — from strategy and prototype to a scalable, secure platform. Let's talk about yours.