AI-Ready Data Platform Accelerating Surgical Innovation
Overview
Carlsmed, a newly public health tech company, needed a unified AI-ready data platform for surgical planning and clinical decision support. ImagineX built a Medallion Architecture on AWS, using Claude Code to accelerate the custom labeling UI, data pipelines, and IaC — delivering the full MVP in ~13 person-weeks against a conventional 20–24 week baseline, with AWS Bedrock powering case summarization via Claude Sonnet and Opus models.
Problem
Disconnected data sources limited AI model development and clinical insights
Lacked secure, compliant de-identification and PHI removal workflows
Manual labeling slowed dataset readiness for ML and LLM use cases
No platform for centralized enrichment, analytics, or clinical review
Needed a fast path to MVP to support executive timelines
Solution
ImagineX built a secure AWS Medallion Architecture with automated ingestion from databases and image repositories. Claude Code accelerated the Next.js labeling UI, Terraform IaC, and Python ETL pipelines — cleaning, deduplicating, and de-identifying data. AWS Bedrock with Claude Sonnet and Opus models powers the Case Summarization product for clinical pattern recognition.
Outcome
Clinical case review time cut 50–65%. Claude-driven summarization of unstructured clinical communications reduced manual case review from ~30 minutes to 10–15 minutes per case, returning clinical team capacity to patient-facing work.
2–3x faster pathology classification. A Claude Code-built Next.js interface streamlined human-in-the-loop labeling and enrichment, accelerating the path to ML-ready datasets while keeping clinical judgment in the loop.
Gold-layer analytics datasets available in hours instead of weeks, compressing Carlsmed's time from raw clinical data to AI-ready insight.
Automated PHI de-identification at scale, validated against HIPAA Safe Harbor criteria, removing manual sanitization as a bottleneck on every downstream analytics and ML workflow.
Production MVP delivered in ~13 person-weeks — a 40–45% reduction against a conventional 20–24 week baseline — hitting executive speed-to-market targets on schedule.
Clinical pattern recognition in production via an AWS Bedrock case summarization service running Claude Sonnet and Opus models on a request-based architecture.
Services
Data & AI Innovation
Enterprise Cloud-Native Engineering
Agile Delivery & Transformation
Technologies Used
Claude Code
AWS Bedrock (Anthropic Claude Sonnet & Opus)
AWS Glue
AWS S3 (Delta Lake)
AWS Step Functions
Next.js
Terraform
Python
