Back office ops · Production

Doctolib rebuilds its data platform into a Unified Healthcare Data Platform for AI and analytics

The problem

Doctolib's centralized, monolithic data platform — built on a single GitHub repository, Airflow instance, and Redshift cluster with shared admin permissions for all users — blocks its ambition to become the leader in AI for healthcare, with CI test runs taking 30–40 minutes, inability to enforce fine-grained access control for sensitive healthcare data, and limited support for event-driven workflows.

Workflow diagram · grounded in source
1
Self-service data ingestion
integration
“Self-Service Ingestion Engine: Empowers teams to ingest data independently with pre-built connectors, validation, and transformation for analytics readiness.”
2
Lakehouse centralized storage
integration
“Lakehouse: Combines the scalability of a data lake with the governance and optimization of a data warehouse, supporting real-time analytics, machine learning, and advanced processing.”
3
ML feature engineering and training
ai_action
“Feature Store: Centralized repository for managing, storing, and serving features used in machine learning models.”
4
LLM operationalization
ai_action
“LLMOps tooling: Provides the infrastructure, workflows, and management capabilities necessary to operationalize large language models (LLMs) in production. This includes tools for model fine-tuning, deployment, monitoring, versioning, pr…”
5
Real-time model serving
output
“Model Serving: Deploys ML models for real-time predictions, managing scaling, versioning, and API endpoints.”
6
Compliance-enforcing transformation
validation
“DataShield Transformer: Enforces security measures like encryption and pseudonymization to simplify data product developers to comply with legal and regulatory standards during transformations.”
Reported outcome

Doctolib is rebuilding its data platform around a Lakehouse, Data Mesh architecture, ML Training Platform, LLMOps tooling, Vector Database, and a compliance-enforcing DataShield Transformer to securely support AI development alongside reporting.
As a prior delivery benchmark, Doctolib completed Tableau Server infrastructure deployment and migration in under three quarters.

Reported metrics
CI pipeline test run time30–40 minutes
Tableau Server migration completion timeunder three quarters
Reported stack
AirflowEKSRedshiftdbt-coreLambdaDynamoDBGitHubPostgreSQLTableau ServerKubernetesCloudflareHL7FHIROMOPDICOM
Source
https://medium.com/doctolib/building-a-unified-healthcare-data-platform-architecture-2bed2aaaf437
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Doctolib is rebuilding its data platform around a Lakehouse, Data Mesh architecture, ML Training Platform, LLMOps tooling, Vector Database, and a compliance-enforcing DataShield Transformer to securely support AI deve…

What tools did this team use?

Airflow, EKS, Redshift, dbt-core, Lambda, DynamoDB, GitHub, PostgreSQL, Tableau Server, Kubernetes.

What results were reported?

CI pipeline test run time: 30–40 minutes; Tableau Server migration completion time: under three quarters (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Self-service data ingestion → Lakehouse centralized storage → ML feature engineering and training → LLM operationalization → Real-time model serving → Compliance-enforcing transformation.

Related back office ops cases