I build data pipelines on production ERP systems. My ML work is academic — and I publish the metrics that don’t flatter it.
I’m Tristán Lacruz, a developer at Minsait (Indra Group). My day job is Dynamics 365 Business Central: APIs, data pipelines, and integrations between ERP, SAP and CRM for purchasing, sales and accounting — processes that fail loudly when the data is wrong. Most of that work turns out to be data quality, which is what pushed me toward machine learning. I’ve written up two projects below as cases: the problem, the data, the decisions, and what I’d do differently. I’m looking for a junior AI/ML engineering role in Germany or Switzerland.
Selected work
Case / academic
Customer Subscription Prediction — the model I threw away
A model scored AUC 0.90. The result of the project is the explanation of why that number was worthless.
- Retracted: 0.90AUC with leakage
- 0.501AUC clean, 5-fold
- ± 0.018Std dev across folds
- 4,000Records
Case / academic
Credit Default Scoring — picking the metric before the model
The model I selected is the one that scored worse on accuracy. Accuracy was the wrong question.
- 0.86AUC-ROC, validation
- 150,000Training records
- 14:1Class imbalance
- 6.7%Default rate
Currently
- Role
- Business Central developer, Minsait (Indra Group) — since March 2025
- Looking for
- Junior AI / ML engineering roles in Germany or Switzerland
- Languages
- Spanish and Catalan native · English C1 (Cambridge) · German B1, working toward B2
- Studying
- BSc Applied Artificial Intelligence, IU Internationale Hochschule — in progress
- Based in
- Valencia, Spain · available to relocate