we're using openai embeddings, chunked 3k chars + an overlap. we're basically using langchain behind the scene so there's no difference