Case Study: Bilt achieves personalized million-agent recommendations with Letta

How Bilt Built a Million-Agent Recommendation System with Letta

Bilt, a housing and neighborhood commerce network, needed a cost-effective and performant way to leverage generative AI for personalized user recommendations. Moving beyond their initial scoring algorithms, the company wanted a system that could creatively understand the nuances of user transactions and engagement to offer truly contextual suggestions. The challenge was integrating vast amounts of user data with large language models without sacrificing creativity or efficiency.

The solution was built using Letta's approach to memory-augmented agents. By processing data into shared memory blocks with powerful models, Letta enabled Bilt to use faster, cheaper models for real-time inference, achieving deep personalization at scale. This Letta-powered system allowed Bilt to rapidly iterate its architecture, ultimately creating over a million tailored agents. The result was a highly effective recommendation engine that internal testers and end-users found remarkably accurate, improving the relevance of offers across their entire platform.

Built

Kosta Krauth

Chief Technology Officer