Thornmind
Artificial intelligence · Open datasets

Better data.
Better minds.

Thornmind designs artificial intelligence and builds the datasets to train it: carefully collected, transparently documented and built to last.

Contribute to the project Learn more
Modelsdesigned and trained
Datasetscollected and verified
Researchshared and open
Mission

An intelligence is only as good as the data it learns from.

Every model carries the choices made about its data. That is why Thornmind brings two sides together in a single project: developing artificial intelligence and collecting the datasets needed to train it. It is a research project with no commercial purpose.

We work so that every piece of data has a clear origin, explicit consent and a precise purpose, so that the models built on it are reliable, verifiable and useful.

01

Collect

Datasets built methodically, from source to documentation.

02

Train

Models developed for research, measured against clear criteria.

03

Share

Results, tools and knowledge given back to the community.

What we do

From data collection to trained model.

Three areas of work that feed into each other.

Datasets

We collect, clean and annotate text, image and audio data, with datasheets that explain their origin, limitations and intended uses.

  • Collection and selection
  • Annotation and validation
  • Full documentation

Models

We design and train artificial intelligence for research, with no commercial purpose, evaluating it rigorously and documenting every result.

  • Architecture design
  • Training and fine-tuning
  • Evaluation and benchmarks

Open research

We publish methods, tools and results, and collaborate on a non-profit basis with anyone who wants to contribute data, skills or ideas.

  • Shared tools
  • Collaborations
  • Publications
Method

A clear flow, from raw data to intelligence.

01

Collection

We identify sources and acquire data with verified consent and licences.

02

Selection

We filter, anonymise and normalise the data to reduce noise and bias.

03

Annotation

We label the data with cross-review and quality checks.

04

Training

We train, evaluate and iterate, documenting every step.

Principles

Building AI you can trust.

The rules behind every Thornmind dataset and model.

Privacy

Respected data

Explicit consent, anonymisation and GDPR compliance.

Transparency

Traceable origin

Every dataset has a datasheet describing its history and limitations.

Quality

Less, but better

We prefer verified data to huge, opaque volumes.

Openness

Shared knowledge

What we learn goes back to the community.

Get involved

Let's build the next generation of data together.

Contribute your data

Do you have data, archives or text collections that could help build better models? Let's work out together how to turn them into a useful, safe and well-documented dataset.

Propose a dataset

Join the team

Researchers, engineers, annotators or simply enthusiasts: there is room for anyone who wants to help shape more responsible AI. Every collaboration is non-profit.

Work with us
Contact

Let's talk.

Write to us with proposals, collaborations or just to find out more.

info@thornmind.it