Zep x LangChain: Diagnosing and Fixing Slow Chatbots

Poor chatbot response times can lead to frustrated users and customer churn. LangChain's new LangSmith service simplifies understanding latency in LLM applications without manual instrumentation. In this article, a poorly performing LangChain app is diagnosed using LangSmith, and performance is improved by an order of magnitude using the Zep memory store. The ConversationSummaryBufferMemory implementation was identified as the cause of latency issues, which were resolved by swapping it out with Zep.