Hi everyone!
I’m working on improving my language model that powers a conversational AI bot. I want to fine-tune it using real interaction data and I’m wondering if LangSmith can help with this.
What I’m trying to do is take the traces from LangSmith that worked well and extract the conversation data from them. Specifically, I need to grab the last model response (from my response agent) along with all the conversation context like system prompts and user messages. Then I want to save this as a JSONL format so I can feed it into Azure AI Foundry for training.
I’ve been looking through the interface but haven’t found a way to export this kind of training data. Does anyone know if this feature exists or if there’s a workaround?
Thanks for any help!