Agentic, conversational and voice AI for member service, designed around value stream owners, human checkpoints and clear escalation paths.
Expertise 01
AI Transformation
AI transformation aligns strategically throughout the enterprise, delivers continuous business value, and increases organizational and operational capability.
I lead AI the way I led Lean: build the operational foundation, govern it, then automate, with people accountable at every level of autonomy. My focus is an enterprise AI and automation capability in a regulated financial institution: one intake, a scored portfolio, NIST-aligned governance and adoption measured in member and team member outcomes.
Build from the base up. Each layer makes the one above it work.
AITechnologyOperational
- 3
Automate with AI and ML, Keeping a Human in the Loop
Predictive models, generative AI assistants and agents, on a process that is already clean. Graduated autonomy: inform, recommend, act with approval, act autonomously.
- 2
Optimize with Technology to Enhance Member Value
Workflow, integration, data platforms and RPA remove the manual steps that remain. Every system change traces back to a member outcome.
- 1
Build the Operational Foundation on Lean Practice
A3 and PDCA, Gemba, process maps, the eight wastes, SOPs and Kanban. Change Agents and a Center of Excellence carry the work.
AI Governance
Governance is an operating discipline, not a binder
Named owners, risk tiers, evidence and stop authority. Govern the system that is running today, not the one that was approved last year.
| Maturity stage | What it looks like | What moves you up |
|---|---|---|
| 1. Ad hoc | Scattered tools, no inventory, rules left to individuals | Sponsor, council, inventory, acceptable-use policy |
| 2. Defined | Policies, tiers and owners exist; reviews happen at approval | Monitoring, re-approval triggers, vendor controls, training |
| 3. Managed | Systems monitored; evidence collected; leadership sees regular reporting | Independent testing, incident exercises, reuse of controls, value tracking |
| 4. Optimized | Continuous assurance, fast approved paths, governance adapts to new technology and rules | Yearly refresh, horizon scanning, lessons learned |
Checklist
AI Governance Checklist for Credit Unions
A high-level checklist organized under the four NIST AI RMF functions (Govern, Map, Measure, Manage), plus credit-union-specific items.
Read the checklistGuide
AI Governance: Best Practices
The role of each team, from the board to risk, compliance, data and the business, and the practices that mature it stage by stage.
Read the guideMachine Learning Models
Models built and explained for the business
Built for the MS in AI in Business at Arizona State University. Each one starts from a business decision, sets its success criterion before training, compares models against a baseline and names its governance limits.
Model 01 · Random Forest regression
Housing Mispricing Model
Random Forest explained 82% of variation in home values (R² 0.821); projected $1.42M a year in avoided overpayment.
At a credit union: Collateral valuation and consistent home-equity and mortgage underwriting.
Read the case studyModel 02 · Decision tree classification
Customer Churn Model
Pruned decision tree flagged at-risk customers, AUC 0.815; tenure and monthly charges were the strongest signals.
At a credit union: Member attrition, dormant-account reactivation and product-depth growth.
Read the case studyModel 03 · Ensemble methods
Credit Risk Scoring
Five models compared; Random Forest ranked applicants best (AUC 0.776), chosen on the cost of each type of error.
At a credit union: Consumer and business lending, credit-limit assignment, with borderline cases routed to an underwriter.
Read the case studyModel 04 · Decision tree classification
Loan Acceptance Model
Pruned decision tree found likely loan takers with 98.1% test accuracy, AUC 0.990.
At a credit union: Next-best-product outreach: deposit to loan, card to personal loan, checking to home equity.
Read the case studyModel 05 · Logistic regression and Streamlit
From Notebook to Business Tool
A loan-eligibility model deployed as a live Streamlit app a non-technical user can run.
At a credit union: Loan-officer decision support, service triage and fast prototypes before engineering investment.
Read the case studyModel 06 · Agentic development
Building Models with Claude Code
An agentic workflow from business question to tested, documented model, with people owning every decision.
At a credit union: Rapid prototyping for an AI intake queue, so engineering capacity goes to what has proven value.
Read the case studyTools: Microsoft Copilot, ChatGPT, Claude Code, Gemini, Python, scikit-learn, Jupyter, Google Colab, Streamlit, MySQL and Tableau.
AI Portfolio
The AI and automation portfolio I govern
Each initiative is managed as a strategic investment, with KPIs defined up front and outcomes tied to member experience, risk reduction and operational efficiency.
AI-assisted origination for lending and deposit accounts, ML-powered lending intelligence and intelligent document processing. AI supports the workflow; credit decisions stay under credit policy.
Behavioral-biometric decisioning that augments fraud review, with model risk, fair-lending and explainability controls designed in from the start.
Uses Microsoft Copilot and ChatGPT daily for analysis, drafting, meeting synthesis and portfolio reporting, and is building standard work and role-based guidance for everyday AI assistants into the Lean Change Agent certification.
Designing the AI Governance Model: intake, risk classification, human-in-the-loop thresholds, data controls, vendor AI due diligence and monitoring, aligned to the NIST AI Risk Management Framework.
White Paper
Designing AI agents that deliver outcomes
Six practices for building an enterprise AI and automation capability in financial services.
One accountable owner, a charter and decision rights, a single intake path and delivery standards.
Score use cases on outcome value, efficiency, risk reduction, feasibility, autonomy tier and strategic fit, and appoint a value stream owner for each member journey (after Mantia, Chatterjee and Lee, 2025).
Automating a flawed process produces faster flaws. Map, baseline and standardize with Lean Six Sigma before an agent touches the work.
RPA, generative AI assistants, agents and machine learning each fit a different job. Decide build, buy or partner on explicit criteria, and keep ownership of the outcome, the logic and the data.
Set an autonomy tier per action, not per agent, anchored in the NIST AI RMF and model risk guidance.
Treat adoption of tools like Microsoft Copilot and ChatGPT as change management, train by role and track KPIs for effectiveness, adoption and business impact.
Read the white paperView the PDF