Modeler
Build models your team can trust and explain.
Built on +30 years of data science expertise, automated into a single click.
Why it’s different
Most AutoML cuts corners. Modeler doesn’t.
Proprietary engine
Multithreaded, vectorized and compiled in C and Assembler. Built by TIMi, not assembled from open-source libraries.
White-box by design
Every variable is explained and every weight is documented.
Thousands of variables
Modeler easily processes datasets with more than fifty thousands columns. No need to pre-filter your data or guess which signals are relevant.
How it works
Five automatic steps, on every run.
Variable selection
Combinations are evaluated in context, not one by one. A model with 8 well-chosen variables often beats one with 80.
5 algorithms competing
Ridge, Lasso, Neural Net, Forest and Trees run on your data. Cross-validation picks the winner.
15,000 models in 5 seconds.
A compiled C and Assembler engine, multithreaded and vectorized, searches the full space of meta-parameters.
4 layers against overfitting
Optimal selection, structural regularization, monotonicity constraints and a temporal stability test.
An auto-generated report
Variables, weights and lift curve, exported to Word, HTML, PNG, SQL, Python, SAS or JavaScript.
Unique to TIMi
The Drift Report: know when a model stops holding.
A model is trained on past data and runs on future data. When that data changes, accuracy degrades quietly. Modeler’s built-in Drift Report watches your models over time.
- Built into Modeler. No separate monitoring tool to buy or integrate.
- Continuous. Model validity is evaluated against the data as it evolves.
- Rarely found elsewhere. Absent from open source, almost absent from all AutoML tools.
In production at SANY. On the engine failure model for mining trucks, the Drift Report identified seasonal model decay, so the model stayed reliable year-round.
Solutions
What it looks like in production.
Churn detection accuracy up from 45% to 57%
MTN analyzed 12.1 billion records across 21.4 million subscribers to catch churn earlier.
€250M in savings kept invested, not withdrawn
An individual-level model showed AXA’s €4.5B savings product was more stable than assumed.
92% forecast accuracy across 300,000 stores
Bavaria runs 30,000 neighborhood models instead of one average for the whole network.
94% of failures predicted, 30 minutes ahead
Predictive maintenance deployed in under a week on a 24/7 aerospace production line.
1,400+ sensors modeled per battery stress test
Volvo isolates which variables degrade EV batteries under thermal shock.
1,800+ infants saved per year
Neonatal complications anticipated up to 6 months ahead, with over 98% precision. 4 people, under 3 months.
Our other tools
Modeler is one engine of the TIMi platform.
Anatella
Data engine
Connects to any source and prepares the dataset Modeler trains on.
Stardust
Segmentation
Builds 3D segmentations directly on the data Anatella has cleaned and joined.
Kibella
Self-service BI
Lets business teams explore the same prepared data, without going back to IT.
Marvim
Data Governance
Connects to data flows and third-party sources to document, trace, and govern your data effortlessly.
FAQ
Common questions.
What is Modeler?
Modeler is TIMi’s predictive engine. It builds and scores predictive models automatically on 100% of a dataset, white-box by design, with a built-in Drift Report to monitor accuracy after deployment.
Is Modeler white-box?
Yes. Every selected variable is explained, every weight is documented, and every decision is traceable back to the data, by design. This is what makes Modeler compliant with the AI Act and GDPR.
How fast is Modeler?
Modeler tests more than 15,000 model and meta-parameter combinations in about 5 seconds, using a compiled, multithreaded C and Assembler engine, then selects the best one mathematically through cross-validation.
What is the Drift Report?
The Drift Report monitors a deployed model continuously and flags when it starts losing accuracy as the underlying data changes. It is a capability almost absent from the rest of the AutoML market and from open-source tools.
How does Modeler prevent overfitting?
Every model goes through four automatic safeguards: optimal selection across the search space, structural regularization, monotonicity constraints aligned with business logic, and a temporal stability test on unseen periods.
Does Modeler need a sample of the data, or sampling at all?
No. Modeler trains up to 100% of the dataset, tens of thousands of variables and millions of rows, without the need of reducing or sampling it first.
See for yourself
Don’t take our word for it.
Run TIMi on your own data with a free trial or request a live demo.
