Dabba: Turning WhatsApp Tiffin Orders into Tomorrow’s Cooking Plan

Описание и применение

Dabba автоматически преобразует сообщения WhatsApp в структурированный список заказов и план приготовления, снижая потери и повышая удовлетворённость клиентов.

Задача

Small tiffin kitchens manually parse WhatsApp group messages to decide tomorrow’s cooking plan, risking waste or loss of customers.

Результат

Dabba achieved 100% valid JSON parsing on 200 held‑out messages, 98% overall accuracy, and generated prep lists in ~30 s, outperforming zero‑shot baselines.

Рабочий процесс

• Paste evening WhatsApp messages
• Parse each message into a structured order
• Resolve relative dates (kal, parso, Monday)
• Apply kitchen memory for customer‑specific phrases
• Forecast final demand per dish
• Display portions to cook with confidence ranges

Доказательства

Dabba converts a WhatsApp conversation into a structured order list and a practical kitchen prep plan.

Источник

The training used three epochs and 129 steps. The total training cost was under two dollars.

Источник

The fine‑tuned 4B model performed much better on the actual task because it had learned the structure of these messages.

Источник

Dabba stores a simple memory in Backboard: POST /assistants/{id}/memories { "content": "Harsh Khan: usual dish paneer_thali, prefs no_onion." }

Источник