From Code to Clinic — Building an Internal Developer Platform for Clinical AI
summary
Institute for AI in Medicine (IKIM) turns clinical questions into a deployed platform: they run real-time ETL into FHIR (over 3 billion data points, new records in ~1 minute), moved from Harvester to Talos, and build an IDP on a BACK-style stack using Backstage, Flux CD, Crossplane and Tekton before replacing compositions with a custom operator. Models are treated as OCI artifacts (model-kit/CNF), with LLM artifact and release gating abstractions, attestations for manual clinical approval, and a working demo that blocks clinical endpoints until approved.
AI-written summary · about
chapters (8)
- 0:00Introduction and mission
- 3:37Code to clinic approach
- 6:10Data integration and standards
- 9:04Kubernetes and infrastructure
- 11:34Internal developer platform
- 18:34Demo LLM platform workflow
- 24:50Takeaways and next steps
- 26:13Q&A session
related talks
33mBridging large-scale FHIR to AI in production
03mAI solutions for healthcare
26mFrom Pixel to Patient: Making Medical AI Work in the Real World
45mAI Digital Twins: From YC Hackathon to Production with Apollo MCP
33mUsing Promptable Image Analysis for Clinical Diagnostics
18mArchitecting Intelligent Platforms
35mBuilding a Medical AI with Kubernetes and Kubeflow
27mGen-AI at Scale: Simplifying Orchestration of Healthcare Applications Across Multi-Cluster Environme