factory IA

Open-source AI for factories, tested before you trust it.

FactoryIA is an independent organisation that builds, tests and validates machine learning and generative AI for manufacturers in Europe and Latin America, using open-source tools only. Every result is published as an academic paper that protects each participant's business value.

Propose a factory use case See how validation works

Validation record

Specimen
Problem
The plant problem, in the plant's words
Model
Name, version and open licence
Data
Where it came from, how much, what was left out
Tested at
Sector, country and line type
Result
Pass, fail or inconclusive, with the numbers
Reproduce
Link to the code and the test protocol

Every project ends in a published academic paper like this one, written to protect this participant's business value.

From a plant problem to a published paper

Most industrial AI stops at the pilot. We run each project through the same four steps so that what comes out can be checked, repeated and reused by another plant.

  1. 1

    Start from the line

    A manufacturer brings a real problem: unplanned stops, scrap, a slow manual inspection.

  2. 2

    Build with open tools

    We assemble a solution from open-source libraries and open-weight models. No licence fees, no lock-in.

  3. 3

    Test on production data

    It runs against data from the line, inside the plant's own environment, with the people who operate it.

  4. 4

    Publish the paper

    Method and results are published as an academic paper, written to protect each participant's business value — including what did not work.

What we put on the test bench

Machine learning

  • Predictive maintenance. Spotting failures from vibration, current and temperature before the line stops.
  • Visual quality inspection. Defect detection with cameras a plant can afford.
  • Process and energy optimisation. Better settings, less scrap, lower consumption per unit.
  • Planning and scheduling. Forecasts and changeover sequences that hold up on a real shift.

Generative AI

  • Maintenance and operator assistants. Answers drawn from your own manuals and work orders.
  • Shift reports and root-cause summaries. Written from line data, checked against it.
  • Work instructions across languages. Portuguese, Spanish, English, French and German.
  • Models that run on site. Open-weight models on local hardware, so plant data stays in the plant.

Open tools we start from

scikit-learn, PyTorch, MLflow, Apache Kafka, Node-RED, open62541 for OPC UA, llama.cpp and open-weight language models. The list changes as tools pass or fail our tests.

Four rules we hold ourselves to

Open source only

If a plant cannot inspect it, modify it and run it without a licence, we do not validate it.

No vendor behind us

We sell no platform. Funding and partners are listed on this site.

Failures get published too

A model that did not work on a real line is a result. It saves the next plant months.

In the plant's language

Records and guides appear in English, Portuguese and Spanish.

Two continents, the same shop-floor problems

A filling line in Paraná and one in Baden-Württemberg stop for the same reasons. What differs is the rules, the funding and the language the documentation is written in. We work across both so a result proven in one can be reused in the other.

Europe

Manufacturers here deploy AI under the EU AI Act. Our records document data, model and test conditions in a form that supports that work.

Brazil

The largest industrial base in Latin America, and our first focus in the region.

Latin America

Spanish-language records and guides for plants from Mexico to Argentina, where open-source references for industry are still scarce.

Take part

You run a plant

Bring a problem and a data sample. We scope a test and tell you plainly whether open-source AI can solve it.

Propose a use case

You build models

Engineers and researchers can contribute code, review a test protocol or submit an open model for validation. Submissions we take on are published and shared with our subscribers.

Submit your open-source solution

You represent an institution

Universities, institutes and industry associations in Europe and Latin America can host a test or co-author a paper.

Become a partner