Martí Villalba
All work

Case study · 2025 · now

Fluxput

Helping technicians find the answer without calling anyone.

Fluxput is an AI platform for field service companies and manufacturers. It turns manuals and catalogues into a copilot that answers on WhatsApp, in the app and on the web, and always says which page each answer comes from.

Role
Co-founder and CPO
Team
4 co-founders
Industry
Field service and manufacturers
Platforms
Web, Android and WhatsApp

The copilot answers on WhatsApp and cites the manual page. Recreation with a fictional model.

01 · The problem

The answer almost always exists. It's in a PDF nobody can read in the middle of a breakdown.

In field service, every question on site ends in a phone call: to a veteran colleague, the office or the manufacturer. The technician loses time, the customer waits and the knowledge stays in a few people's heads.

~70%

of repeated technical questions were already answered in the manual the technician had on their phone.

Our own field analysis, 2026.
The technician

Standing in front of the machine with their hands full. They need a short, reliable answer.

The service company's office

Assigns the work orders and takes the calls when something gets stuck.

The manufacturer

Writes the manual, then answers the questions the manual already covered.

02 · Decisions

Three decisions that shaped the product

  1. WhatsApp before yet another app

    Technicians already live on WhatsApp. Putting the copilot there removed the friction of installing and learning something new.

    Trade-offLess control over the interface: the design is made of text, lists and attachments, with no screens of our own.

  2. Always cite the source, and know how to say “I don't know”

    During a breakdown, a made-up answer is worse than none. Trust is designed: every answer shows where it comes from.

    Trade-offSome questions go unanswered. We prefer that to an answer that sounds right and isn't.

  3. One knowledge base per model, not one PDF

    Organising the documentation by unit makes answers precise and lets the manufacturer see what gets asked about each product.

    Trade-offMore work when onboarding each catalogue.

03 · The copilot

How it answers: short, with the page, and nothing made up

Every answer has three parts: what to do, which manual and page it comes from, and what happens if the information isn't documented. In that case the copilot says so and suggests checking with the manufacturer. The work order lives in the field app, where the technician closes the job report with photos, audio and the customer's signature.

OT-2481In progress

X200 heat pump service

Check high-pressure switch
Clean heat exchanger
Measure operating pressure
Job report photos and audio
Close and sign
F
Technical copilotonline
Tightening torque for the 4-way valve?
That figure isn't in the documentation for this model. I'd recommend checking with the manufacturer.
⚑ No source · doesn't make things up
And the wiring diagram?
Here it is, section 6.2.
▤ X200 manual · p. 112

Work order in the app and copilot with the source cited. Illustrative screens with a fictional model.

04 · My role

From discovery to release

Discovery

Interviews and visits to service companies and manufacturers to understand where time gets lost.

Copilot design

Conversation flow, source citations and behaviour when the answer isn't documented.

Field app

Work orders, job reports with photos and audio, and the customer's digital signature.

Roadmap and releases

Prioritisation, specs and shipping versions up to 3.6, with the WhatsApp assistant.

05 · Results

What moved

5

modules in production: knowledge, copilot, analytics, operations and WhatsApp.

3.6

version in production, with the WhatsApp assistant answering from the catalogues.

2026

Fluxput, certified as a startup by ENISA.

What I learnedThe person who uses the product isn't always the one who pays for it. Technicians use Fluxput; manufacturers gain the most when their manual answers on its own.