portfolio / pilot product Pilot
Voice-to-quote assistant for woodworkers
The woodworker sent a voice message on WhatsApp and got the quote back as a PDF. The pilot showed where a language model cannot work alone.

The problem
A self-employed woodworker writes quotes by hand and takes too long to answer the customer. The idea was to dictate customer, material and deadline and get the document ready.
What I did
- I was the technical lead: official WhatsApp API, n8n workflow, audio transcription and PDF generation.
- The model extracted the data from speech and also did the math for the quote.
Result
The pilot ran with real woodworkers. What blocked the product was the model getting the math wrong. In the end, the code and the accesses were handed over to the client.
What I learned
The model interprets and the code calculates. Today I would do the extraction with structured output and leave all the math to a tested function.
Stack
- Official WhatsApp API
- n8n
- LLM