Hiflylabs migrates AI application to Databricks for unified security and governance

Hiflylabs developed an AI application that transforms unstructured medical data from various sources (lab results, medical imagery, EHR, etc.) into a 360 patient view accessible to doctors via an interactive web interface. The app allows doctors to query medical data with AI chat and generate patient summaries and insights in clicks. 

The application was built on Azure OpenAI and relied on hand-picked LLMs. The team set a goal to enhance governance, data velocity, and AI compatibility by rebuilding the app on one unified platform. The migration had to adhere to strict compliance and security standards and enable fully governed, trustworthy AI—without extensive custom development or performance tradeoffs.
 

Hiflylabs revamped app UX and rebuilt the application entirely on Databricks under a month. By combining a unique first-mover experience with Databricks Apps, stress-tested scalability strategy, and built-in Databricks tools, the team delivered a secure, trustworthy healthcare application without months of setups, approvals, and handovers.

• Databricks Apps serves as the foundation, allowing the team to rebuild an AI application using original backend/frontend frameworks without added complexity and development overhead.
• Built-in robust security, authentication, and governance eliminate the need for additional security tools or policies, cutting weeks of custom development and approvals. Access control and user management within Databricks Apps ensure doctors only see their own patients, while AI agents operate with proper data isolation to prevent clinical data mix-ups.
• Unified governance across the entire platform with Unity Catalog provides automatic data lineage, tracking, and audit that address sensitive healthcare data requirements—all operating within one environment for improved data flow, safety, and traceability.
• Explainable AI architecture with citation-backed responses mitigates hallucinations and helps build clinician trust. Agentic RAG with intent classification retrieves relevant data and provides answers with citations and links to sources, making every AI-generated insight traceable and verifiable.
• Databricks-langchain expands LLM options and simplifies integrations with LLMs regardless of their hosting.
• Real-time LLM endpoints serve GenAI content generation.
• Model Serving hosts custom predictive models/agents for reasoning, red-flag alerts, and patient insight generation.
• Lakebase handles transactional and analytics data while acting as a low-latency cache for conversations, dashboard loading, and key-value storage of patient data.
• Vector Search with Unity Catalog improves data management and velocity.
• Painless CI/CD integration with GitHub actions and Databricks Asset Bundles enables one-click deployment with built-in monitoring and observability.
• Proprietary scalability strategy combining serving endpoints and Jobs enables automated scaling and management of heavy-compute tasks.
• A custom feature to scale down resources when unused provides cloud cost control.
• Revamped UX with new agentic AI generates insights, identifies red flags, and makes predictions to support triage and speed up decision-making for doctors.

The unified platform approach meets both current clinical needs and future growth requirements without compromising performance. Everything operates within one environment, providing reliable, audit-ready data management, security, and traceability.