Solution
Bayer developed E.L.Y. Crop Protection (Mini), a domain-specialized small language model built on Microsoft Phi and hosted in Azure AI Foundry. The system ingests proprietary agronomic product label data and regulatory rules, fine-tuned to deliver precise crop protection guidance without hallucination. Frontline agricultural advisors query the model through E.L.Y. Copilot, which retrieves label-accurate answers in real time. Azure AI Foundry provides infrastructure for model versioning, compliance logging, and source-linked traceability. The solution maintains full IP ownership, audit trails, and rapid update cycles to incorporate label changes and regulatory shifts without retraining. Strategic retail partners access the model through secure APIs and access controls, enabling scaled deployment across the United States with measurable productivity gains.
Data flow
Proprietary product label data and regulatory rules are fine-tuned into the Microsoft Phi model within Azure AI Foundry. When advisors query E.L.Y. Copilot with crop protection questions, the model retrieves contextual label information, processes through the fine-tuned Phi architecture, and returns compliant, label-accurate answers sourced to specific documents. All responses are audit-logged and version-controlled in Azure AI Foundry, enabling rapid updates when label changes or regulatory shifts occur without retraining from scratch.
Solution architecture
4 components · 4 layers - Compute
- Microsoft Phi Small language model foundation fine-tuned exclusively on proprietary agronomic product label data and regulatory rules.
- Storage
- Azure AI Foundry Hosts the domain-specialized model with version control, compliance logging, and IP ownership for rapid iteration and updates.
- Serving
- Azure AI Foundry Models Provides the catalog and publishing infrastructure for sharing the fine-tuned model with strategic agricultural retailer partners.
- Uncategorized
- Microsoft Cloud for Industry Provides compliance architecture guidance and managed services to ensure the solution meets regulated agricultural use case requirements.
- Aligned to Microsoft's solution accelerator for regulated AI deployments
- Fine-tuned on proprietary Bayer label data and regulatory rules
- Hosted in Azure AI Foundry Models with versioning and IP control
- Small language model on Microsoft Phi architecture
- Source-linked, audit-logged responses for compliance
- Surfaced via existing E.L.Y. Copilot interface and exposed to partners through secure APIs
Azure AI Foundry also gives us IP ownership and lets us control updates, access, and traceability.
The model is hosted in Azure AI Foundry Models, allowing Bayer to retain full IP control while enabling rapid iteration, versioning, and traceability.
We estimate 5-10% productivity gains among early users. Complex questions that once took days or weeks to resolve now take under 30 seconds.
We fine-tuned Microsoft Phi using our internal datasets. We worked closely with Microsoft’s Cloud for Industry team to shape the model’s architecture and ensure it aligned with our compliance and operational needs.
