Case Study: Evolve cuts sleep coach memory misses in half with Langfuse

Evolve cuts cross-session memory misses in half with Langfuse

Evolve, a YC-backed company, built Rest, a voice-and-text AI sleep coach based on CBT-I principles to address the scarcity of accessible insomnia treatment. Their challenge was to trace and evaluate this complex voice agent system, which managed over 100 dynamically composed prompts, and to identify and fix issues like the coach's inconsistency in remembering user context across sessions.

Using Langfuse for end-to-end tracing and prompt management, Evolve gained a centralized platform for error analysis and monitoring. Langfuse enabled them to implement LLM-as-a-judge evaluators to measure cross-session memory, which led to a project that cut the coach's memory misses in half in its first iterations. The solution also provided a safe way for non-engineers to manage and deploy prompt changes without touching code.