Workflow Automation &
Decision Support
​​
I don't build "chatbots." I build systems that remove manual work and improve decision consistency.
​
The most valuable AI applications serve as a "second set of eyes"—technology that augments human experts rather than replacing them. In 2025, the enterprise AI market reached $18 billion with intelligent process automation leading adoption. Companies are moving beyond pilot projects to production systems that deliver measurable productivity gains.
​
My focus is on AI that integrates into existing workflows, reduces cognitive load, and provides decision support where expertise matters most. Whether it's pre-highlighting anomalies for review, summarizing contextual data, or flagging inconsistencies, the goal is always the same: give experts back their time while improving outcomes.
Prototype → Pilot → Scale
Speed matters. I prototype fast to validate value, but I scale with "platform discipline"—evals, guardrails, and observability.
​
The 2024-2025 AI landscape taught us that deployment maturity separates successful AI programs from perpetual pilots. Modern workflow automation demands:
​
Rapid Prototyping: Tools like no-code AI platforms enable business users to test concepts in days, not months. This acceleration has cut automation turnaround time by 40% across leading enterprises.
​
Disciplined Scale: Production systems require evaluation frameworks, drift monitoring, and human-in-the-loop validation. The companies winning in AI aren't deploying faster models—they're deploying better infrastructure.
​
Platform Thinking: Reusable components, API-first architecture, and continuous feedback loops. This is how you go from one successful automation to ten, then a hundred.
High-Stakes Environments
Experience in Healthcare (IBM Watson) and Enterprise Data (Oracle/Datto). I understand compliance, data provenance, and the cost of hallucinations.
​
In regulated industries, "good enough" isn't good enough. Healthcare, financial services, and enterprise systems require:
​
-
Data Governance: Clear lineage, audit trails, and privacy compliance
-
Validation Protocols: Continuous evaluation against ground truth
-
Explainability: Stakeholders need to understand why the system made its recommendation
-
Graceful Degradation: Systems must fail safely and transparently
​
My background spans medical AI deployment at IBM Watson Health and managing 4.7 million device telemetry streams at scale. I've seen what happens when systems fail in production—and built the guardrails to prevent it.
Case Study: AI in Radiology
Solving the "Polar Bear in a Snowstorm" Problem at IBM Watson Health
The Solution: We didn't try to replace the radiologist. We built a "second set of eyes." An AI advisor to highlight suspicious regions, summarize patient history, and surface relevant priors—reducing cognitive load without demanding blind trust.
​
The design principle was augmentation, not automation. Radiologists responded to tools that quietly pulled forward context they needed but rarely had time to assemble manually.
​
The "Builder" Reality: Success wasn't about showmanship; it was about the "plumbing." Data quality, governance, integration into actual workflows, and continuous validation. Watson Health's broader story taught the industry that clinical impact requires less hype and more infrastructure.
​
The lesson I carried forward: great AI products start with understanding the failure mode, designing for the human in the loop, and building the data backbone first.
​
The Outcome: By 2020, Google Health's mammography AI (deployed through iCAD) demonstrated the maturity of this approach—reducing false positives by 5.7% and false negatives by 9.4% in production use. The technology moved from concept to concrete clinical impact.
​​
​
"Clinical impact requires less showmanship and more plumbing—data quality, governance, validation, and UX integration."
The "Builder PM" Discipline
Case Study: AI in Radiology
Solving the "Polar Bear in a Snowstorm" Problem at IBM Watson Health
Case Study: AI in Radiology
Solving the "Polar Bear in a Snowstorm" Problem at IBM Watson Health
The AI Product Manager role has evolved. In 2025, success requires more than understanding models—it requires business operator discipline.
​
Augmentation, Not Replacement
Design AI to keep the human in the loop for high-value decisions. Automation works best when it enhances expertise rather than attempting to replace it.
​
The Boring Parts Matter
Data provenance, labeling policy, bias audits, and drift monitoring are not optional features; they are the product. Companies spending $7.3 billion annually on departmental AI have learned this lesson—infrastructure investment separates pilots from production.
​
ROI Focus
My background as a business operator (private lending, self-storage) means I evaluate AI through the lens of P&L and operational efficiency, not just technological capability.
​
Real Example: At Datto, our predictive device failure system reduced support tickets by 38% while cutting cloud costs by 40%. The AI model mattered—but the serverless data architecture and cost optimization mattered more.
Experience Includes
Built AI-powered workflow automation at IBM Watson Health (Healthcare)
Led data infrastructure modernization processing 3.2TB daily (SaaS at scale)
Managed 14 mobile product releases for enterprise construction platform (Oracle)
Delivering product in complex, regulated ecosystems.