Technology I bring: hands-on stories

Technical leadership at a glance

1. Building an AI-native care platform myself (CareTogether, 2026)

The challenge

Build a production platform for seniors, families, and care teams with a three-person founding team, handling health information safely from day one.

What I built

• Application: Next.js, React, and TypeScript in a monorepo, deployed on Google Cloud Run, with scheduled jobs on Cloud Scheduler.
• Data: PostgreSQL on Cloud SQL over a private network, and Cloud Storage encrypted with customer-managed keys.
• Identity and privacy: Firebase Auth with one-time text codes, data loss prevention that redacts sensitive details from images, and text alerts that carry no health details.
• AI: voice check-ins and LLM agents grounded through retrieval in each member's care plan, medications, prior conversations, and open tasks; a second model reviews every call.
• Integration: bidirectional EHR integration through FHIR R4 and HL7 v2; payments, voice, and messaging through vendor APIs.

1. Check-inThe senior answers a phone call and talks in their own words
2. ContextThe agent retrieves the care plan, medications, history, and open tasks
3. SafetyGuardrails, crisis routing, and a second model reviewing every call
4. SignalMissed medications, falls, changes in routine, low mood, ranked by priority
5. ActionFamily, care team, or clinician; EHR updated; a person handles anything sensitive

The point

The hard product questions in an AI system are what context to retrieve, what the model may act on, and when to hand off to a person. The data, not the model, is the source of truth.

2. Shipping AI-written code without losing control (CareTogether)

The challenge

AI coding tools make a small team fast, but speed without discipline creates code nobody can maintain or audit.

1. ChangeAI-assisted code on a branch
2. EvidencePull request with risk tier, tests, and rollback plan
3. ChecksAutomated tests and an independent AI review
4. ApprovalHuman sign-off for health data, permissions, and billing
5. ReleaseDeployed with a full audit trail

What I built

• Every change is a pull request with evidence: what changed and why, a risk tier, test output, and a rollback plan.
• Continuous integration on every pull request: type checks, linting, automated tests, and end-to-end browser tests.
• An independent AI code review on every change, separate from the AI that wrote it.
• Risk tiers: interface changes merge after tests pass; logic changes need review; anything touching health data, permissions, billing, crisis paths, or database migrations needs my approval and a soak period.
• A full audit trail from each change to its tests, review, and approver.

3. From on-premise hardware to a cloud platform (iCAD)

The challenge

iCAD's AI shipped on physical servers installed at each site. Deployment averaged about 43 days, and the company had to procure, build, ship, and repair hardware worldwide.

What we built

• A cloud platform with Python and C#/.NET services in Docker on Google Kubernetes Engine, infrastructure as code with Terraform, CI/CD pipelines, and cloud monitoring, with defined uptime targets and disaster recovery.
• Integration into hospital systems through DICOM, HL7, and FHIR: 70+ PACS vendors and 30+ imaging device models from competing manufacturers.
• Microsoft Fabric on Azure for analytics, alongside Google Cloud and AWS.

The result

• Customers: live in about 3 hours instead of 43 days, with no servers to host and fewer support issues.
• Reliability and speed: 99.99% uptime, and releases went from once every 6 to 12 months to monthly, reaching every site at once.
• Operations: no hardware to buy, build, ship, or repair, which lowered field and implementation costs.
• Business: the move made subscriptions possible: ProFound Cloud grew from launch to $10M+ ARR in under two years, and company ARR grew 78%.

4. An AI data pipeline built for regulatory review (iCAD)

The challenge

Regulated AI is only as good as its data, and every training image needs a traceable history.

What we built

• A training-data pipeline grown from about 7M to 20M+ mammography images from 100+ clinical sites, with full provenance.
• Natural language processing that extracted findings from radiology and pathology reports, and linked each patient's exams over time.
• Labeling, quality control, training, validation, and production monitoring, run by a data science and machine learning team of about 15, with releases only when models met defined performance targets.
• Deep learning (convolutional neural networks) and machine learning models, and an evaluation of a multimodal foundation model against task-specific models through the Google partnership.

The result

iCAD's first FDA AI clearance in four years.

5. A consented health data platform (Susan G. Komen)

The challenge

Bring each patient's records from many providers into one place, with consent, and keep patient health data separate from donor data.

What we built

• An event-driven data platform on Azure that ingested digital and paper medical records, claims, and genomic data through FHIR and a data broker's APIs, normalized them to a canonical model aligned with USCDI, and applied natural language processing to unstructured records.
• Salesforce Health Cloud for the patient portal and navigator console; Power Platform, Power BI, and Azure AI Search for analytics and search.
• A rules-based next-best-action engine on Azure, pushing recommendations into Health Cloud.
• Consent checks on every flow into the research environment, with patient health data kept separate from donor data.
• I led Komen's engineering team, which built the patient care center platform, and partnered with IQVIA, which built the EHR integration, the registry, and ShareForCures. I owned the product strategy, requirements, and architecture decisions.

The result

Consolidating siloed systems cut response times 80% and saved $1M+ a year, and the platform powered the patient care center and the ShareForCures research registry.

6. Running enterprise IT and security (Komen and iCAD)

• Ran all of IT at both organizations: cloud migration, single sign-on, Azure security, and data loss prevention.
• Built security and privacy programs to NIST and HIPAA at Komen, and to ISO 27001 and GDPR at iCAD, with GDPR at Komen as well.
• Implemented and ran NetSuite ERP and Salesforce at Komen; ran Microsoft Dynamics ERP at iCAD, including procurement, manufacturing, and returns.
• Owned the payment experience and fraud controls for $200M+ a year in donations, and kept digital properties compliant with Section 508 and WCAG accessibility standards.

7. Modernizing a SaaS platform and adding machine learning (SigmaFlow)

• Replaced and wrapped legacy SOAP services with event-driven microservices on AWS (Lambda, API Gateway, SQS, and EKS), moving a multi-tenant platform to cloud SaaS without disrupting customers.
• Connected the platform to operators' SCADA systems and built optimization and simulation models for what-if planning.
• Ingested sensor data on AWS to train machine learning models that ranked the next best action for each asset, with feature-level explanations.
• Owned uptime targets and disaster recovery for the platform. The company grew revenue 10x to an exit.

8. Where it started

I began as a software engineer at Thomson Reuters, Abbott Laboratories, MicrograFX, and Igenesis. At Ignite Technologies, I grew from senior architect to head of product architecture on .NET platforms, then VP of Product across four product lines.

Skills

References

• iCAD teams up with Microsoft (MD+DI, April 2025)
• IPKeys acquires SigmaFlow (Washington Technology, February 2019)
• Identity and access management at Susan G. Komen (BizTech, November 2020), citing my comments on single sign-on and password policy
• iCAD launches ProFound Cloud, built with Google Cloud (iCAD, April 2024)
• iCAD expands its agreement with Google Health (iCAD, August 2023)
• iCAD collaborates with Microsoft on AI-powered mammography solutions (iCAD, April 2025)