Creating a Secure and Agile ML Platform Using Amazon SageMaker with Qred

Qred offers small and midsize enterprises an alternative to large traditional banks. The company operates in seven countries, providing business loans, credit cards, and an invoicing service. According to a 2023 Financial Times report on Europe’s 1,000 fastest-growing companies, Qred is the fastest-growing company in Sweden’s fintech industry.

The company wants to maintain that momentum while scaling to additional markets. With more than 10 data scientists working on ML models remotely and autonomously, Qred needed to increase efficiency so that it could continue to meet demands and regulations while growing. Qred was already AWS native and started exploring Amazon SageMaker in 2022. Instead of building and training models using local environments, Qred’s data scientists are now using Amazon SageMaker to train and manage ML models in production.

Qred handles sensitive customer information, and the company received its banking license in 2023, which requires additional layers of security. Qred enhances security by storing data centrally. “With a centralized platform using Amazon SageMaker, compliance is simpler,” says Lezgin Bakircioglu, chief technology officer at Qred. “It’s simpler to add sensitive data when we have it centralized and secured.”

Using SageMaker Projects, which help organizations set up and standardize developer environments, Qred substantially increased efficiency. Qred reduced the time required to set up each model from 8 hours to 1 hour, significantly supporting the company’s growth goals in view of the multitude of models that it has operating across various markets.

Qred’s data scientists benefit from streamlined workflows and reduced complexity around tooling, access, and infrastructure. The employee onboarding process is also simpler. As data continues to accumulate, Qred automates model retraining using Amazon SageMaker, enhancing accuracy and saving time.

Using Amazon SageMaker, Qred gains features that expedite results generation. Qred’s comprehensive model training pipeline with tracking and versioning facilitates simpler troubleshooting. With a standardized approach, data scientists can work on another team member’s model if they are out. Qred can now train larger models on diverse datasets in parallel or conduct parallel training of multiple neural networks with the same data.