Skip to main content

Astellas Employees Build 420+ Custom AI Models on a Secure, Self-Hosted Platform with Open WebUI

Astellas Pharma Inc.

Overview​

info

How Astellas Pharma Inc. built a secure, flexible internal AI platform with Open WebUI on Azure, enabling 3,000+ employees to create, customize, and share generative AI models. Users organically built 420+ custom models and reported a +43 NPS across the organization.

At a Glance​

  • Users: 3,200+ total; 800+ advanced users with access to frontier models
  • Region: Global (headquartered in Tokyo, Japan)
  • Industry: Pharmaceutical
  • Deployment: Azure AKS (private endpoints)
  • Models: Azure OpenAI, Gemini, DeepSeek, Perplexity
  • Time-to-deploy: ~1 month (April–May 2025)
  • Adoption: 30–40% weekly active users sustained over five months
  • Key Results: 420+ custom models created, 68% of users report significant efficacy gains, +43 NPS

About Astellas Pharma Inc.​

Astellas Pharma Inc. is a global life sciences company engaged in the research & development, sales & marketing, and manufacturing of prescription drugs. With operations in roughly 70 countries and nearly 20,000 employees and contractors, Astellas is driven by a mission to turn innovative science into VALUE for patients worldwide.

The Challenge: Flexible, Secure AI Without Vendor Lock-In​

As generative AI capabilities advanced rapidly, Astellas needed a platform that could keep pace, without tying the organization to a single model provider. The company operates across dozens of countries, manages complex regulatory landscapes, and supports scientific workflows that span multiple languages and disciplines. Third-party SaaS tools offered limited integration options, restricted customization, and lacked the security and compliance controls a global pharmaceutical company requires.

The goal: deploy a secure, self-hosted AI platform that gave every department, from R&D and Clinical Development to Pharmacovigilance and Compliance, the flexibility to leverage cutting-edge models while maintaining full control over security, compliance, and data governance.

Key Requirements

  • Flexible model selection: freedom to choose and switch between providers
  • Option to use locally-hosted models for sensitive analysis
  • Capability to create custom models tailored to specific workflows
  • Secure sharing and joint development of models within defined groups
  • Technical integration with internal and external systems

The Solution: Open WebUI on Azure​

Open WebUI was selected for its flexibility, fine-grained permission controls, and scalable architecture, allowing Astellas to host entirely within their secured private environment while supporting both custom model creation and group-based collaboration.

Architecture Highlights

  • Compute / Orchestration: Kubernetes (AKS) cluster with autoscaling based on demand
  • Storage / Database: Azure Database for PostgreSQL (encrypted at rest; backups aligned to internal policy)
  • Networking: VNet-isolated; Application Gateway with private endpoints for LLM APIs
  • Secrets: Azure Key Vault integrated with Open WebUI for secure credential management
  • CI/CD: Azure DevOps pipelines for upgrades, configuration deployment, and governance controls
  • Logging / Monitoring: API Management Service, Azure Dashboards, Databricks
  • Security Controls: MFA via IdP, RBAC by group, data residency enforced

β€œOpen WebUI allowed us to create and share custom AI models securely across the entire company, while giving us the flexibility to leverage the full potential of any cutting-edge model available.” - Generative AI Team Manager, Astellas

Models & Data Handling​

Astellas runs a multi-model, multi-provider strategy that gives teams the right tool for every task:

  • Primary: Azure OpenAI models
  • Research augmentation: Gemini, DeepSeek, Perplexity (web search augmentation)
  • Model selection: Advanced users can select from all available models; beginners are offered curated options for ease of use. Default models are assigned by user group with manual switching available
  • RAG: Embedding models and chunk sizes optimized for performance. Sources include SOPs, regulatory policy documents, and external medical research data
  • Sensitive data handling: A deliberate governance-first approach. All users complete training on confidential data handling before access; usage is monitored for compliance; access to shared knowledge and models is governed by operational process design that enforces least-privilege principles

Adoption & Enablement​

Astellas took a comprehensive approach to enablement, running a two-month onboarding program for 2,000+ employees that covered demonstrations, interactive workshops, and hands-on sessions. Organization-wide enablement was completed by the end of the rollout period.

Training included:

  • Interactive sessions on generative AI fundamentals
  • Open WebUI operations training covering model creation and sharing
  • Ongoing office hours for advanced use and Q&A

Over the following months:

  • Weekly active users stabilized at 30–40% over five months
  • Users organically created hundreds of custom models, sharing them across departments
  • All departments adopted the platform, including Research, Clinical Development, Medical, Sales, Marketing, Legal, Compliance, Pharmacovigilance, Administration, Communications, and Corporate Strategy

β€œOur efficiency in gathering external scientific information has improved dramatically. Being able to select and switch models depending on the use case makes our research far more effective.” - Research Department User, Astellas

Results: Democratization, Productivity, and Satisfaction​

Screenshot of Open WebUI showing the Astellas internal AI platform.

01. User-Driven Model Creation​

Employees created custom generative AI models tailored to their specific workflows, from research brief templates to regulatory document analysis, without waiting on IT. The platform now hosts 420+ user-built models, a clear signal of organic, bottom-up adoption.

02. Organic Growth & Adoption​

The platform grew to 3,200+ total users organically, with an advanced user base exceeding 800, demonstrating strong demand-driven adoption across the organization.

03. Measurable Efficiency Gains​

68% of users reported significant efficacy gains in their daily work, with a +43 NPS indicating strong user satisfaction across departments.

04. Research Acceleration​

R&D and research teams reported dramatic improvements in gathering and synthesizing scientific information, clinical trial summaries, and multilingual medical documents.

β€œFor coding tasks, efficiency has increased more than ten-fold, I can’t imagine working without this tool now.” - Advanced User, Astellas

Top Use Cases​

  • Drafting and reviewing scientific research briefs for R&D projects
  • Summarizing clinical trial reports
  • RAG over SOPs and regulatory policy documents
  • External information gathering on medical research data
  • Translation of multilingual medical documents
  • Generating code snippets for data analysis pipelines in research

Integrations​

  • Microsoft Outlook: AI-assisted drafting and scheduling of responses
  • Microsoft Teams: Collaborative AI-powered chat for project discussions
  • SharePoint: Secure document ingestion and search within AI workflows
  • Internal document management system: Instant retrieval of corporate knowledge base

Why Astellas Chose Open WebUI​

  • Model freedom: choose, switch, and update models without vendor lock-in
  • Custom model creation: employees build and share models tailored to their workflows
  • Fine-grained permissions: RBAC and group-based collaboration controls
  • Self-hosted flexibility: fully private deployment within Azure
  • Deep integration: connects to existing Microsoft ecosystem and internal systems
  • Scalable architecture: grew from initial deployment to 3,200+ users organically
tip

Looking for an Enterprise Plan? β€” Speak with Our Sales Team Today!

Get enhanced capabilities, including custom theming and branding, Service Level Agreement (SLA) support, Long-Term Support (LTS) versions, and more!

This content is for informational purposes only and does not constitute a warranty, guarantee, or contractual commitment. Open WebUI is provided "as is." See your license for applicable terms.