PYTHON · LLMSHIPPED
WhatsApp To-Do Extractor
A tool that reads a WhatsApp conversation and extracts the actual to-dos buried in it — using LLM prompts, and working across English, Hindi, and Hinglish messages rather than assuming everyone types in clean English.
The hard part: which tasks, and whose
Extracting "things that sound like tasks" is easy and useless. The real work was context-aware classification:
- Distinguishing a task someone assigned to themselves from one assigned to someone else
- Strict deduplication, so the same commitment mentioned three times across a thread becomes one to-do, not three
From side project to production
The approach worked well enough that it directly inspired the production Promises feature at Cloop.ai — a Sthaan.ai product — for real-time WhatsApp to-do extraction. A good reminder that a sharp little experiment can become the seed of a shipped product.
Stack: Python · LLM APIs · Regex.