Prumo

Prumo · AI for furniture and metalwork factories · Technology by ArtorIA

AI proposes.
People confirm.

How I designed the way a 100-person factory and its AI work together: the floor reports by voice note, Prumo drafts the update, and every important change waits for a person’s yes.

Prumo voice reports: a WhatsApp voice note, what the AI understood and the change waiting for a person
  • ~10,000 m²first client's factory
  • 100+people on the operation
  • WhatsAppthe only tool the floor needs
  • Near-dailyfeedback cycles with the client

The problem

The owner found out about a delay once it was already a loss.

Every project moves through cutting, metalwork, painting, glass, varnish and assembly, each with different people, documents and suppliers. The information lived in three places:

  • WhatsApp chats
  • Spreadsheets
  • People’s heads

The founder had a working prototype, but no design system and screens where everything competed for attention. And there was one hard constraint: factory workers don’t want a new system.

Who it serves

Four people, four questions

  • Leadership

    “What's at risk, and does the real cost still fit the budget?”

  • Purchasing

    “What do I need to quote, and who's holding up the project?”

  • Designers

    “Does the 3D match the 2D, and where is this part?”

  • Factory floor

    “How do I report my work without learning a new system?”

01

Where should I look first?

The projects screen answers in seconds. It’s a to-do list for the owner’s attention.

Projects screen with status chips, risk sorting and summary numbers
  1. Status in plain words

    “At risk · glass” tells you what's wrong and where. No code to decode, no chart to read.

  2. Risk first, by default

    The list sorts itself by what needs attention, so the owner doesn't have to go looking.

  3. Four numbers, no more

    Projects running, deliveries this month, projects at risk and approved value. The rest is one click away.

  4. Starting a project is one drop

    Send the 3D or the drawing and the engine plans the stages for you to review.

02

What am I building, and what will it cost?

One project, one screen: the model, the drawing, the parts and the budget stay together.

Project detail with the 3D model, 2D drawing, parts list and budget button
  1. 3D, 2D or both

    One switch compares the views, and the 3D gets the most space.

  2. One selection, everywhere

    Click a part and its row lights up, with size and price.

  3. Numbers with context

    Each part shows its cost right where you’re looking.

  4. The project's vital signs

    Yield, area, weight and days, generated from the model.

  5. The budget starts here

    The riskiest flow now runs inside the project, from the 3D.

03

What happened on the floor today?

Workers keep doing what they already do: they send a WhatsApp voice note. The AI does the rest, and a person has the last word.

Voice reports with a transcript, the AI summary and the change to the timeline
  1. Strangers don't get through

    An unknown number changes nothing until someone identifies it.

  2. Confidence, field by field

    You see what the AI is sure of: who, which project, which stage.

  3. The proof stays attached

    The audio and transcript sit right above the summary.

  4. Preview before it happens

    You see what changes on the timeline before anything moves.

  5. One click to say yes

    Confirming takes a second. Reviewing should never feel like work.

Principles

Four rules behind every screen

  1. 1One screen, one question

    How much does it cost, where does it stand, who is late. Everything else is one click away.

  2. 2AI proposes, people confirm

    AI does the heavy lifting. Every important change passes through a person.

  3. 3Numbers come with context

    Each value sits next to its budget, deadline or percentage, so nobody does the math.

  4. 4Color only means status

    Green, yellow and red are never decoration, so they keep their meaning.

Where it stands

From loose features to a product that tells the story of every project.

Prumo is still in development, so there are no usage numbers yet, and I didn’t have direct access to users: I worked from the founder, the client’s profile and the existing prototypes, with feedback almost every day.

What I delivered: a continuous flow from the 3D project to delivery, the budget brought inside the project, a way to feed the system that doesn’t change the floor’s routine, and a design system the founder now uses to code the next screens.

The lesson I took with me: in AI products, trust is a design decision. It has to be drawn on the screen.

Want to see more?

Get in touch with meSee other projects