Platform Engineering
Turning clinical NLP models into a scalable, API-ready product
How Savana built a Google Cloud-native architecture to expose NLP models via API and enable scalable external consumption.
Savana is a healthtech SaaS company delivering scientifically validated, multilingual clinical NLP trained exclusively on electronic health records, enabling the generation of deep real-world evidence for hospitals, HTA bodies, pharmaceutical companies, and MedTech.
Website
Industry
Healthcare Technology & Clinical Data
Location
Madrid, Spain
Stack
Google Cloud Platform · Vertex AI · Cloud Run · API Gateway · Cloud Build · BigQuery
Overview
Productionisation of core NLP models in Google Cloud Vertex AI.
API-based access to clinical NLP models for on-demand processing.
Standardized model deployment and inference using Vertex AI, with centralized model registry and consistent serving logic.
The challenge
From service to product via AI productization
Savana had already created a powerful high-throughput data processing pipeline for large-scale inference with clinical AI models, but was lacking a product via which clients could directly interact with those models for real-time inference.
Productizing these models required a new architecture, with standardized deployment, API access and operational control.
This made the move to Google Cloud, who introduced a fully integrated data and AI stack, a strategic step. The project focused on building a new architecture to support a pilot for selected models, rather than replacing the existing setup.
Our process
Turning clinical NLP models into a scalable, API-ready product
01
Building a Google Cloud-native AI foundation
Astrafy designed and implemented a new architecture on Google Cloud to host, manage and serve Savana’s NLP models. This included setting up a standardized model deployment and inference using Vertex AI, with model registry and consistent serving logic, creating a solid and scalable foundation for current and future models.
02
Productizing models through an API layer
To enable external usage, Astrafy built a fully operational API layer. Through API Gateway and Cloud Run, clients can now send clinical documents and receive structured outputs in real time.
The solution includes authentication, quotas, monitoring and logging, turning internal models into a consumable service.
03
Enabling scalability, distribution and autonomy
Beyond infrastructure, the architecture was designed for long-term use and growth. Astrafy delivered full documentation, architecture guidelines and enablement, allowing Savana’s team to continue building independently.
The solution is also aligned with Google Cloud standards, preparing it for distribution via the Marketplace.
Results
Turning AI into a scalable product
What began as an internal, service-based use of clinical NLP models has now become a scalable product.
Savana's models are accessible via API for external consumption, allowing clients to send clinical documents and receive structured outputs in real time. This shift also unlocked a new monetization model, with usage-based API access turning what were once internal tools into a revenue-generating service.
Beyond the immediate product, the architecture was built for long-term use and growth, aligned with Google Cloud standards. This positions Savana not only for future scale, but also for distribution through the Google Cloud Marketplace.
Latest case studies
