When companies deploy their agents into production, a key challenge emerges: how to evaluate whether the agent is performing as expected. You might find yourself asking: Is my agent on the right track? How can I ensure the final output is accurate?
In this video, we walk through how to build and evaluate a customer support agent, covering:
- The challenges of evaluating agents and practical approaches to overcome them
- How to create a golden dataset to evaluate against
- Evaluation strategies to assess agent performance
When watching, we recommend you follow along with the code in the notebook below. Have questions? Comment below or join us in the LangChain Community Slack: https://www.langchain.com/join-community
Resources:
- Notebook Link: https://github.com/langchain-ai/the-judge/blob/main/build-eval-agent/agent-eval.ipynb
- LangGraph Docs: https://langchain-ai.github.io/langgraph/tutorials/introduction/
- LangSmith Docs: https://docs.smith.langchain.com
- LangGraph Studio: https://langchain-ai.github.io/langgraph/concepts/langgraph_studio/
- Running evaluations with the LangSmith SDK: https://docs.smith.langchain.com/evaluation
- Enroll in LangChain Academy for free with our Introduction to LangGraph and LangSmith courses: https://academy.langchain.com
- Learn more about ReAct: https://langchain-ai.github.io/langgraph/how-tos/create-react-agent/