Wisper

The problem

A heating and photovoltaic technician spends the day on site with gloves on and both hands busy. Then the report has to be filed: client, job, description, hours, kilometres, expenses.

The interesting part is when it happens. Back at the office, near closing time, at the end of a day that started at seven, by someone who is tired and wants to go home. It is a form, filled in by hand, one field after another, and nothing about it is automatic. So it gets rushed, or written from memory two days later, or forgotten, and the office spends the next morning chasing it.

This is not a hypothetical user. It is the daily routine at my father's technical firm, where I build software for industrial clients, and where I have been watching this problem from close up for the past few months.

How it works

1. Say “Wisper”

   The wake word runs entirely on the phone. Until it is heard, no audio leaves the device and no connection is needed.

2. Just talk

   The fields fill themselves in on screen while you speak. Correct anything out loud, “no, three hours”, and it updates.

3. Say “confermo”

   Wisper reads the report back, you confirm, and the row lands in the company Google Sheet. The phone is never touched.

The app speaks Italian, because that is who it was built for. The English under each line is a translation, not something the app says.

On site, it goes first

1. It knows where you are

   Open it standing in a client's yard and it opens the conversation itself: “I see you are at Rossi's, and the only job open here is the Fara Vicentino photovoltaic. Is that right?” It proposes, it never assumes; you might be at the office writing up yesterday.

2. It tells jobs apart the way you do

   Ask what is open for a client and it answers by place, not by code, because a technician says “the one in via Roma” and never “M004”.

3. It knows when to shut up

   A wake word means holding the microphone open forever, and on Android the microphone belongs to whoever grabbed it. So Wisper listens when the screen is off, or when it is the app you are looking at. Use the phone for anything else, or take a call, and it steps aside on its own.

4. It keeps the sheet tidy

   New clients and new jobs get created by voice, and it says out loud when it does it. Say the work is finished and the job closes in the company sheet, so tomorrow it is not offered again.

What it is not, yet

This was built in a weekend hackathon. These limits are real and known, and are listed here rather than discovered later.

• Single user. There is no login and no accounts: the app writes to one spreadsheet, the owner's. A production version needs authentication, one workspace per company, and permissions for each technician.
• The spreadsheet endpoint is unprotected. The bridge to Google Sheets is a public Apps Script URL. Anyone holding it could write a row. In production it needs a real token.
• It needs a connection to understand you. The wake word is fully offline, but the reasoning and the save are not. On a site with no signal, Wisper wakes up but cannot yet queue the report for later. That queue is the next thing to build.
• Noise degrades accuracy. Published Italian speech recognition benchmarks, around 3% word error, are measured on clean studio audio. On a real building site, expect meaningfully worse. This is the honest weak point of the whole idea.
• There is no APK to download, on purpose. The model key would travel inside it and anyone could pull it back out. Gemini keys cannot be locked to an Android package the way Maps keys can, so the only real fix is a server holding the key and the app asking that server. That server is not built yet, so the project is source only and you build it with your own key.

Where it goes

A spreadsheet is where this starts, not where it ends. The next step is writing into the system a firm already runs, an ERP like Odoo, so the report lands as a timesheet entry against the right job instead of a row somebody copies over by hand.

• Signatures. The surveyor's is signed once and kept in Drive. The client signs on the phone there and then, offered only when the job is being closed, because a signature on unfinished work means nothing.
• Receipts, read by the phone. Photograph it, OCR it, attach it to the expense. Right now an expense is a number somebody said out loud and the accountant takes it on trust.
• Then the plumbing. Offline queue with automatic retry, one account per technician, a real token on the spreadsheet endpoint, a server holding the model key.