Synthetic bilingual (Indonesian 79% / English ~21%) SFT dataset for a WhatsApp
cake-shop chatbot with 9 LangChain tools. Goal: sharpen tool-calling
accuracy on a small model without degrading conversational ability
(63% tool-call rows : ~37% conversation/refusal/clarification rows).
Built for the production system it serves — the system prompt, tool schemas
(generated from the live code via convert_to_openai_tool), and… See the full description on the dataset page:
https://huggingface.co/datasets/LasagnaS/toti-cakery-toolcall.
