I write product contracts teams can build, instrument, and evaluate—then make the tradeoffs inspectable in a working prototype.
I work where product strategy, experimentation, analytics, and trustworthy AI meet. My portfolio is built for the questions strong product teams ask: What is the real friction? What are we deliberately not building? What evidence would change the decision? What breaks at the edge?
Portfolio simulation: Fictionalized company context and synthetic scenario data. The PRD, decision logic, acceptance criteria, instrumentation plan, and working prototype are my original work.
A cancellation experience designed to earn the next renewal—not trap the current one.
Live decision lab · Redacted PRD · Friction map · Experiment plan
- Frames cancellation completion and post-flow trust as non-negotiable guardrails.
- Includes implementation-level acceptance criteria, dependency failure behavior, analytics definitions, and stopping rules.
- Ships with a responsive, interactive decision dashboard using clearly labeled synthetic scenarios.
Portfolio simulation: Synthetic conversations, policies, experiment inputs, and results. The product contract, experiment analysis, decision logic, and working prototype are my original work.
An AI support copilot that knows when not to send.
Live product + readout · Experiment readout · PRD-lite · Friction map
- Tests targeted evidence gates for high-risk AI-generated support claims.
- Shows validity checks, effect sizes, guardrails, segment heterogeneity, limitations, and what would reverse the rollout decision.
- Turns the aggregate result into a bounded launch: ship where evidence is strong, hold where it is not.
Operating loop: Problem → Friction → Hypothesis → Product contract → Build → Measure → Decision
- Write for action. Non-goals stop scope drift; acceptance criteria remove interpretation at handoff.
- Treat metrics as contracts. I define the population, numerator, denominator, guardrails, and decision threshold before reading results.
- Use AI with boundaries. Confidence is not proof. I design for provenance, human judgment, failure recovery, and auditability.
- Build to learn. A working prototype exposes tradeoffs that a polished deck can hide.
| Project | Product problem | What it demonstrates |
|---|---|---|
| Signal to Roadmap | Turn fragmented customer signals into an evidence-backed roadmap | Full-stack AI product, synthesis, challenge mode |
| PriorityLens | Make feature prioritization explainable and auditable | Decision systems, bias detection, product strategy |
| ScopeCreep | Detect requirement drift before timelines collapse | NLP, workflow design, delivery risk |
| Smart Onboarding Analyzer | Diagnose activation drop-off and size interventions | Growth, funnels, prioritization, ARR scenarios |
Product: PRDs · Product strategy · User research · Friction mapping · Acceptance criteria · Roadmapping · Growth loops · Activation & retention
Experimentation: Hypothesis design · A/B testing · Guardrails · Metric contracts · Cohorts · Decision readouts
AI products: LLM UX · Retrieval and evidence · Evals · Confidence calibration · Human-in-the-loop systems · Safety and auditability
Builder stack: Python · FastAPI · TypeScript · Next.js · Streamlit · SQL · scikit-learn · Pandas · Plotly
- Open to product builder, AI product, product strategy, growth, and technical product roles
The best product work makes the decision—and its tradeoffs—easy to inspect.