Zep x LangSmith: Foundations of LLM app development with LangChain.js and Zep

The blog post is a tutorial on how to build three foundational Language Learning Model (LLM) apps using TypeScript, LangChain.js, and Zep. The first app is a simple conversational bot that recalls past conversations. The second app is a Q&A over Docs/RAG-type app that uses Zep's VectorStore to support a `ConversationalRetrievalQAChain` searching over a Zep document Collection. The third app is a REACT-type agent that uses Zep's conversational history and vector store as tools. The post also highlights the use of LangSmith, a platform for observability, providing insight into what the chains and agents are doing under the hood.

The tutorial also discusses the importance of thoughtful chunking and data preparation for the performance of the apps. It also explains how Zep's features, including message metadata, can be maximized when customizing or building your own agents and tools. The post concludes by encouraging readers to sign up for the LangSmith beta and get set up with Zep using the Quick Start Guide.

Key takeaways

Zep and LangChain.js can be used to build three foundational LLM apps: a simple conversational bot, a Retrieval Augmented Generation app, and a REACT-type agent.
Zep's long-term memory store simplifies the process of adding relevant documents, chat history memory, and rich user data to prompts.
Zep's VectorStore can be used to support a ConversationalRetrievalQAChain searching over a Zep document Collection.
LangSmith platform provides insight into what chains and agents are doing under the hood, improving visibility into the data sent to the LLM.

Zep x LangSmith: Foundations of LLM app development with LangChain.js and Zep

Key takeaways

Discussion (0)