Travel policy
§ 2 · Rail travelTrain travel is reimbursed in standard class. First class requires prior approval.
ALLY AI / LAB
Train a real small model and explore how sources make answers traceable. Everything uses example data right in your browser.
Start exploring ↓THE ALLY AI LAB
How does a model learn? How do documents become an answer you can trace? Two small windows into big possibilities.
Real training · in your browser
Start training. A small neural network learns to tell two groups apart. The coloured area shows how it would classify new points.
Ready. Start training and watch the decision area change.
This example shows how a neural network learns from examples to tell two groups apart. With JavaScript enabled, you can train the model here and test its predictions.
How would the model classify a new point?
A small network with two inputs, two layers of twelve neurons each, and one output adjusts its weights through backpropagation. Adam optimises log loss. Data and starting weights are reproducible. Everything runs locally. Accuracy refers to these 120 training points, not unseen data. This learning example is not a language model.
Current log loss:
Prepared example · fictional documents
Choose a question. See which passage answers it – and when the available information is not enough.
Choose an example question
Standard class is reimbursed. First class requires prior approval.
Source 01 · Travel policy, § 2Train travel is reimbursed in standard class. First class requires prior approval.
Purchases above €1,000 require approval from both the budget owner and Finance.
New colleagues receive their IT equipment on their first day. The team lead requests access rights before the start date.
The answers are editorially prepared, not generated live by AI. This example illustrates source-based answers. A real application needs suitable data, access controls and quality checks.