Decision-Centered Product Systems
I work on products where customers, operators, evidence, workflow, data, business rules, technology, and human judgment have to remain coherent enough for people to make consequential decisions.
My work connects discovery, product direction, workflow and decision architecture, information design, technical prototyping, implementation, and human–AI interaction. The goal is not simply to add features. It is to understand the operating system around the product well enough to decide what should change, test it, and carry that intent through delivery.
Start with the decision, not the interface.
In complex products, the visible screen is often the last step in a much larger system. A person may be trying to decide what to do while the product is also reconciling policy, account state, evidence, risk, eligibility, prior actions, technical constraints, and organizational responsibility.
I begin by understanding the decision itself: who is making it, what information they need, what rules or uncertainty shape it, what happens next, and what the system must make visible for the decision to be understandable and accountable.
Make product complexity visible before committing to the solution.
Product complexity accumulates across workflows, business rules, data, roles, channels, reusable components, handoffs, exceptions, and legacy systems. When those relationships remain implicit, teams can optimize a feature while making the larger system harder to operate.
I use research, contextual inquiry, workflow maps, service structures, prototypes, information models, working code, and end-to-end scenarios to make those relationships concrete enough for customers, product leaders, designers, engineers, analysts, and operators to challenge together.
Keep product direction connected to what actually ships.
Strategy loses value when it becomes detached from technical and operational reality. I stay close to implementation so emerging product direction can be tested against actual behavior, data, reusable architecture, edge cases, and engineering constraints while decisions are still inexpensive to change.
That creates a shorter feedback loop between what customers and users need, what the organization intends, what the technology can support, and what the delivered product actually does.
I work directly with customers, end users, business owners, technical teams, analysts, and program leaders to understand how the product and operating environment actually behave before defining a direction.
Interviews, contextual inquiry, workflow analysis, prototypes, analytics, and implementation evidence become inputs to product decisions: which problem matters, where friction originates, which constraints are real, what can be simplified, and what should be tested before larger investment.
I map how requests, information, rules, approvals, exceptions, handoffs, and decisions move across people and systems. This exposes where responsibility is unclear, where users are compensating for product gaps, and where automation or workflow changes can remove friction without hiding accountability.
The resulting structures give product and engineering teams a common model for deciding what belongs in the interface, what belongs in the workflow, what should be automated, and where human judgment remains necessary.
Data becomes useful when people can understand what changed, what matters, where the information came from, and what action is available. I use information architecture, visualization, status models, prioritization, and interaction design to bring operational context into the moment a decision is made.
The goal is not more reporting. It is better product behavior around evidence, state, uncertainty, tradeoffs, and next actions.
AI changes a product decision when a model begins interpreting evidence, generating recommendations, retrieving policy, prioritizing work, or suggesting an action. I design the human-facing system around that capability: purpose, evidence, provenance, retrieval behavior, output constraints, review gates, escalation, source visibility, and human control.
I am particularly interested in the boundary between automation and judgment: what the system should decide, what a person needs to understand before acting, what remains uncertain, and where accountability belongs.
I use working prototypes and direct implementation to test feasibility, expose hidden assumptions, and shorten the distance between product intent and delivered behavior.
Depending on the system, that has included Angular, AngularJS, NativeScript, HTML/CSS/JavaScript, Qlik, Power BI, R Shiny, Flask, Databricks, reusable component libraries, live data, and production code. The technology is not the point; keeping product decisions inspectable through delivery is.
At Softwise, I led UX architecture and system design across five financial applications while working directly in the Angular production environment. The work spanned loan origination, point-of-sale, servicing, refinancing, payments, collections, title lending, and administration.
I worked directly with business owners, end users, and developers to connect business requirements, user workflows, data behavior, interaction logic, and technical constraints into a coherent product architecture. Rather than treating each workflow or branded client as a separate product problem, we used shared components and reusable product behavior so changes could remain coherent across applications, channels, and client-specific shells.
Borrowers had to move through identity, application, eligibility, disclosures, financial information, validation, and next-step decisions without needing to understand the internal systems behind them.
I designed step structures, validation and error-recovery patterns, help-in-context, information hierarchy, and reusable components so financial rules, data collection, borrower understanding, and implementation behavior pointed toward the same path.
Staff and partners were guiding customers through financial products in real time while simultaneously managing product rules, disclosures, required information, eligibility, and system state. The product had to support compliance without pulling attention away from the customer conversation.
I structured point-of-sale and title-loan flows, tuned field order and branching logic, reduced unnecessary steps, and embedded prompts and guardrails for required disclosures. The resulting paths required fewer keystrokes and less backtracking while keeping product terms, risk rules, customer-facing guidance, and required steps aligned.
Collections and support agents were working across multiple systems to understand account status, payment history, promises to pay, prior contacts, and risk signals. Customers were often already under stress, so agents needed the full story before asking the first question.
I designed a decision-support console around the real conversations agents had every day: account context, information hierarchy, prioritization rules, action flows, and scripting aids in one usable workflow. The design aligned servicing data, collections policy, customer history, and agent judgment so agents could work more fairly, consistently, and quickly with a clearer view of the customer's situation and the next best action.
I facilitated Air Force stakeholders around document assessment, reassessment, approvals, signatures, and mobile approvers, mapping roles, handoffs, and movement across paper, PDF, and digital systems to define a workable end-to-end service process.
Related Air Force AI and data work included Qlik dashboards, R Shiny prototypes, NLP across accounting data, RAG-modeled assistant workflows, and high-fidelity HTML/live-data prototypes used with stakeholders to test assumptions and refine requirements. The common pattern was to make complex behavior concrete enough for customers and technical teams to inspect, challenge, and improve together.
I managed the client experience for an NLP initiative and helped deliver a retrieval-augmented generation project, implementing fixes directly when expected and actual system behavior diverged.
My current independent work extends that pattern through custom GPT and AI-agent systems built around knowledge structure, retrieval, evidence, provenance, decision logic, output constraints, human review, escalation, and source visibility. I test for drift, conflicting evidence, edge cases, uncertainty, and non-linear interaction paths so machine assistance remains inspectable rather than becoming an opaque replacement for judgment.