CTO  ·  AI Governance  ·  Data Platform at Scale

AI programs fail
not because of the models.
Because of what comes after.

I help engineering teams build the governance layer, data infrastructure, and architectural discipline that makes AI defensible, auditable, and production-ready at scale.

500M+
Devices at peak scale
$2B+
Annual business impact
15yr
Building at Apple scale
Thinking
Most teams are one hard question away from a serious problem.

Enterprise customers are getting sharper. They ask how your AI systems make decisions, what happens when a model is wrong, and whether your data practices survive a GDPR audit. Most engineering teams cannot answer these questions. Not because they lack talent. Because they were never asked to build for that standard.

The AI governance gap is an architecture problem that compounds silently every week you ship without addressing it. By the time it surfaces in a deal room or a board meeting, the remediation cost is three times higher. The teams that build governance early use it as a competitive advantage — faster enterprise sales cycles, fewer production incidents, higher customer trust.

I spent 10 years at Apple where the bar was not just "does it work" but "can we defend every decision this system makes, to anyone, at any time." That standard shapes every system I architect and every engagement I take on.

Expertise
Four domains. One integrated system.
01
AI Governance and Zero Trust Architecture
Model registries, explainability layers, audit trails, drift detection, rollback playbooks. ZTA applied to AI infrastructure. Not advisory — built and shipped in production at Apple scale.
Deployed Apple-wide
02
Data Platform at Petabyte Scale
End-to-end data architecture from ingestion to governance. Delta Lake, Kafka, Spark Streaming, Databricks, Snowflake. Built for the volume where most platforms break and governance disappears.
10TB/day  ·  500M+ devices
03
Privacy Engineering and Compliance
GDPR, CCPA, HIPAA, SOC 2, EU AI Act. Privacy-by-design at the architecture level. Led Apple App Tracking Transparency — one of the most consequential privacy launches in consumer tech history.
ATT  ·  GDPR  ·  HIPAA
04
Agentic AI and ML Infrastructure
End-to-end ML pipelines, on-device inference, RAG orchestration, agentic workflow systems. From sensor-level data instrumentation through model deployment, monitoring, and automated retraining.
95%+ accuracy  ·  On-device
Work
Selected case studies.
Real systems. Real numbers.
Case 01
Data Intelligence Platform — enabling AI/ML across Apple at petabyte scale
Apple Inc  ·  2011-2019
Data PlatformAI/ML InfrastructurePrivacy EngineeringPetabyte Scale
[collapse]
The Problem
Apple had no unified framework for collecting, governing, and analyzing device telemetry at scale. Engineering teams built ad hoc solutions in silos. There was no way to understand device behavior across 500M+ devices, train ML models on reliable data, or defend data practices under legal scrutiny. Product decisions were made without confidence. Legal teams had no foundation to defend the company in litigation.
Devices instrumented500M+
Daily data volume10TB / day
Metrics served2,500+
Teams served500+
Annual impact$2B+
Budget$50M / 5yr
What Was Built
Designed and built from zero — an analytics framework running on every Apple device. On-device data collection with OTA configuration management. User consent and opt-in controls. Privacy-by-design with anonymization and differential privacy baked in from day one. The same platform served Legal teams defending lawsuits, Product teams making launch decisions, AppleCare diagnosing devices in store, and AI/ML teams training on production-quality labeled data.
Design Principle
Governance is not a constraint on the data system. It is the foundation that makes data trustworthy enough to act on. Without consent controls and privacy architecture built in at the start, none of the downstream AI/ML use cases would have been legally or operationally possible.
The Outcome
Grew from a single metric to 2,500+ metrics across 500+ teams because it was trusted, not mandated. Informed the decision to remove the Ethernet port from MacBook — a billion-dollar hardware call made with data from 500M+ devices. Powered Fall Detection ML, FaceID telemetry, and Apple Health studies. Never caused a single critical incident or blocked an iOS, MacOS, or WatchOS release in its entire operational lifetime.
Case 02
Fall and Crash Detection — life-safety ML on-device
Apple Inc  ·  WatchOS / iOS
On-device MLTemporal Convolutional NetworkMLOps PipelinePrivacy-by-design
[expand]
The Problem
Detect a fall within 200ms on Apple Watch and trigger an e-SOS call. Detect a vehicle crash on iPhone and do the same. Everything had to run on-device, offline, with near-zero latency, near-zero false positives, and negligible power consumption — because a missed call could cost a life. Every model and data decision had to pass privacy review.
Detection accuracy95%+
Latency targetUnder 200ms
OperationOnline and Offline
ModelTCN via CoreML
Budget$15M / 3yr
What Was Built
Selected Temporal Convolutional Networks after evaluating model families for power efficiency, latency, and long-range dependency requirements. Redesigned on-device analytics with a ring-buffer memory model capturing millisecond-level accelerometer and motion data. Built end-to-end MLOps: Kafka ingestion, Spark Streaming normalization, PySpark feature engineering, model training orchestration, evaluation gating, canary deployment, and drift monitoring. Every pipeline stage embedded consent boundaries and audit logging.
Design Principle
In safety-critical ML, governance at the pipeline level is as important as the model architecture. Feature engineering consent controls, training data audit trails, and drift detection protocols were non-negotiable design constraints, not afterthoughts.
The Outcome
Delivered a TCN model with 95%+ detection accuracy, reduced false positives, and negligible battery impact. Fully operational offline — critical for emergencies without connectivity. The MLOps pipeline and ring-buffer architecture became a reusable foundation for subsequent motion-driven AI features across iOS and WatchOS.
Case 03
Unified AI and Data Governance — enterprise transformation at Lucid
Lucid (Ad-Tech, Acquired)  ·  2022-2023
Enterprise GovernanceZero Trust ArchitectureDelta LakeAI/ML Enablement
[expand]
The Problem
Post-acquisition: 60+ business units with incompatible data ownership models, no lineage tracking, and active GDPR/CCPA conflicts. A CDC pipeline powering all BI reporting had a silent data corruption issue causing undetected revenue loss. Major clients — Disney, Netflix, Amazon — were demanding AI-powered measurement the platform could not support.
Revenue enabled$300M+
AWS reduction25% ($4M/yr)
Domains aligned60+
AI data growth70%
Pipeline savings$500K/yr
Measurement revenue+30%
What Was Built
Hybrid federated architecture with centralized governance via Collibra and Unity Catalog, regional data planes preserving compliance. Migrated Redshift to Delta Lake on Databricks. Deployed Zero Trust Architecture with Okta. Organized a company-wide Data+AI Summit to secure buy-in from 60+ domain leaders — because governance without adoption is shelfware. Resolved the CDC corruption by discovering 70% of upstream data came from API calls, then enforcing schema validation at source.
Design Principle
Enterprise AI governance is 20% architecture and 80% organizational change. The technical solution and the change management strategy were co-designed. Domain leaders who helped build the framework felt accountable to it rather than constrained by it.
The Outcome
Resolved the revenue leak saving $500K annually. Grew actionable AI data by 70% enabling fraud detection and brand-lift measurement for Disney, Netflix, and Amazon Ads. Reduced AWS costs by $4M/yr. Measurement revenue grew 30%. $5M secured from C-suite for the multi-year AI strategy.
Case 04
Agentic AI Orchestration — healthcare enrollment at AimCare
AimCare  ·  2023-Present
Agentic AIRAG / LangChainHIPAA / SOC 2Multi-tenant SaaS
[expand]
The Problem
Healthcare enrollment is a multi-step, document-heavy process requiring significant manual provider time. The platform had to be HIPAA-compliant, SOC 2 certified, integrated with EHR systems via FHIR/HL7/X12, and capable of complex eligibility logic — built from zero with a lean team on a startup budget.
Provider workload reduction40%
ComplianceHIPAA + SOC 2
EHR integrationEPIC / FHIR / HL7
Build timelineUnder 12 months
Market size$1B+
What Was Built
Multi-tenant B2B SaaS with full EHR integration. Six-agent AI orchestration — Eligibility, Application, Document, Consent, Task, and Notification Agents — over gRPC. RAG-based AI using LangChain, OpenAI, and Gemini for health plan Q&A. Zero Trust Architecture with per-tenant network isolation. Each agent has explicit consent boundaries, audit logging, and escalation paths embedded in its decision logic.
Design Principle
In regulated industries, compliance cannot be a wrapper around agentic AI. It must be woven into each agent's decision logic, audit trail, and escalation path. Governance and orchestration are the same architectural layer.
The Outcome
Full platform delivered in under 12 months. Provider enrollment workload reduced by 40%. HIPAA and SOC 2 readiness via Vanta and Drata. Hospital network pilots secured. AWS, Vanta, and NVIDIA partnerships established. Agentic orchestration became the primary technical differentiator in acquisition conversations.
Leadership Philosophy
How I lead. How I think. How I build.

My leadership model is built around four principles I call VIEW. Each is a deliberate choice about where to invest energy. Together they produce teams that are self-managing, technically excellent, and built to outlast my involvement.

V
Vision
Long arcs, short building blocks
I
Integrity
Culture where truth surfaces early
E
Empowerment
Ownership over supervision
W
Warrior
Fearless on hard problems
Set the long arc.
Make today a building block.

The most common failure in engineering leadership is short-termism dressed as urgency. Speed matters. But shipping without a clear destination creates the technical debt and governance gaps that kill velocity twelve months later and cost three times as much to fix.

My approach is to set an ambitious 18-36 month vision and break it into 90-day building blocks that each deliver standalone value. Every increment should be a proof point, not just a checkpoint. At Apple, the Data Intelligence Platform was a 5-year vision built in annual releases that each stood alone. At Lucid, the governance transformation was a 3-year commitment delivered through quarterly wins that kept 60+ domain leaders bought in.

The vision earns patience from stakeholders. The building blocks earn trust from the team. You need both.

Build the culture where
problems surface before they are expensive.

Integrity in engineering leadership is about building systems and culture where the truth is visible before it becomes a crisis. The most dangerous signal in any organization is a leader who only hears good news.

Psychological safety is the highest-leverage investment a leader can make. When engineers feel safe surfacing problems, bad technical decisions get caught in design review instead of production incidents. When managers raise resource conflicts early, they get resolved in planning rather than quarterly reviews.

I use data as an integrity tool — not surveillance, but visibility. A KPI turning red on a Monday scorecard is a gift. At Lucid, we discovered a silent revenue leak in the CDC pipeline precisely because we had the right metrics reviewed weekly. Without that visibility, it would have continued for quarters.

Build self-managed teams.
Make yourself unnecessary.

The goal of a Director or CTO is not to be the smartest person in the room. It is to build a room where the team makes good decisions without needing you in it. That requires resisting the pull toward involvement, pushing ownership downward, and measuring success by how little firefighting you do.

I structure organizations around clear ownership boundaries, automated visibility, and lightweight governance. Monthly delivery reviews with automated KPIs for cost, latency, and adoption. Not micromanagement — organizational infrastructure that makes accountability inherent rather than enforced.

When someone on my team is underperforming, my first question is not "why are they not working harder." It is "what constraint am I not seeing?" Most underperformance comes from being spread too thin or unclear on priorities. Remove the constraint first. Assess capability second.

Fearless on hard problems.
Disciplined about how to solve them.

The warrior principle is about a specific relationship with difficulty. Hard problems — ambiguous requirements, political resistance, impossible timelines, inherited technical debt — are not obstacles to the work. They are the work.

At Apple, we were asked whether removing the Ethernet port from MacBook would affect users. No existing telemetry answered this directly. Rather than declaring it infeasible, we designed a custom PySpark analysis of network throughput patterns across the global Mac population. The result: 70% of users primarily used Wi-Fi. A billion-dollar product decision made with statistical confidence.

The warrior mindset also means staying consistent under pressure: governance standards, architectural discipline, and not taking shortcuts that create future crises. The discipline to build it right the first time is the most underrated competitive advantage in engineering.

Experience
15 years building.
Every role a harder problem.
2023 - Present
Chief Technology and Product Officer
AimCare — Health AI/Tech
Co-founded AI-powered healthcare enrollment platform. Agentic AI orchestration system. HIPAA/SOC 2 compliant. AWS, Vanta, NVIDIA partnerships. Hospital network pilots secured. Positioned for acquisition.
2022 - 2023
Director of Engineering — Data Platform, AI/ML Infra, Cloud Ops, Privacy and Governance
Lucid (Ad-Tech, Acquired)
Led 45+ engineer org. Unified data governance across 60+ domains. ZTA with Okta, Snowflake, Databricks, Collibra. $4M/yr AWS reduction. Clients: Disney, Netflix, Amazon. $5M budget secured from C-suite.
2021 - 2022
Privacy Engineering Leader — Governance and Mobile Ecosystems
Apple Inc — Apple Privacy
Led 20 privacy engineers. Owned launch of App Tracking Transparency. GDPR/CCPA/HIPAA alignment across 100+ product teams. Privacy-by-design embedded at the architecture level.
2019 - 2020
Sr. Engineering Program Manager — Global Infrastructure
Apple Inc — Apple Maps
Directed 100+ engineer org. Petabyte-scale data lakes and real-time streaming pipelines. Maps Search, EV Routing, CarPlay. 25% release cycle efficiency improvement.
2011 - 2019
Data and Platform Engineering Lead — AI, MLOps, Analytics
Apple Inc — Apple Diagnostics and Analytics
Built the Data Intelligence Platform from zero. Fall/Crash Detection ML. Exposure Notification. 2,500+ metrics across 500+ teams. $2B+ annual business impact.
Education
MS Computer Science — Purdue University
Executive Leadership — UC Berkeley  ·  Executive Program — Stanford University
The Offer
One entry point.
Three ways to engage.
// Start here
AI Readiness Audit
Two-week structured assessment across five domains: strategy, infrastructure, model risk, privacy architecture, and team capability. Written report and prioritized roadmap delivered regardless of whether you proceed to a retainer.
$8,500
flat fee  ·  2 weeks
credited toward month one if you proceed
Starter
$15,000
per month  ·  90-day minimum
2 days per week embedded
Governance charter and ownership model
Architecture review
Team coaching
Bi-weekly executive reporting
Most Popular
Core
$25,000
per month  ·  90-day minimum
3 days per week embedded
All Starter deliverables
Vendor evaluation and ZTA integration
Data platform governance
Model risk and drift monitoring
Board-level reporting support
Strategic
$40,000
per month  ·  90-day minimum
CTO seat
All Core deliverables
Board-level reporting
Hiring and vendor negotiation
Strategic roadmap ownership
Full AI program ownership

Maximum 3 concurrent clients  ·  Outcomes defined upfront  ·  15% rate increase after first term

// ready when you are
Schedule a call.
No pitch. No commitment.

Use the form to pick a time that works. We will spend 30 minutes understanding where your AI program is today and whether an AI Readiness Audit is the right next step.

// Book a 30-min call
Pick a time that works for you