Selected Work

AI products,
made clearer.

I write prompts, structure knowledge, diagnose workflows, design QA, and advise product teams on the experience around the model. These case studies show the decisions and artifacts behind that work.

Start Here

A support system that had reached its limit.

This case study shows the full shape of my work: language, knowledge, evaluation, product decisions, and rollout across a live AI system.

Case Study · Enterprise AI

Scaling support AI across 8 markets

A rule-based support bot had reached its limits across 20M+ users. I led the move to grounded, evaluated LLM support with prompt architecture, output verification, and staged rollout.

Read the support AI case study

The evidence

01
Find the system limitContainment plateaued and unsupported responses remained.
02
Design the language layerPrompts, tools, grounding, verification, and specialist roles.
03
Measure the change25% faster responses, 60% fewer unsupported responses.

Role Lead Conversation Designer and AI Product Manager.

Outcome About $2M in estimated annual savings.

25% faster responsesMeasured against the prior support experience.

40% CSAT liftCustomer satisfaction improved after the migration.

60% fewer unsupported responsesGrounding and verification raised the trust floor.

200+ operators enabledThe system was introduced across the organization.

More case studies

The work behind the message.

Each case study shows the problem, my role, the artifacts I shaped, and the evidence available.

Workflows

Find where customers lose the thread.

Case Study2022

Self-Service Workflow Redesign

Rebuilt self-service around how customers actually find help, make decisions, and move through a connected-fitness product.

Faster resolutionclearer product feedback

Customer journey
Information architecture

Read the case study

Conversations

Make language useful to people and systems.

Case Study2024

Rebuilding a 50K-conversation intent taxonomy

Studied real customer language and reorganized the intent structure so routing, reporting, and human handoffs reflected the needs behind the conversations.

Clearer routingreporting · handoffs

Intent architecture
NLU content

Read the case study

Case Study2025

Migrating support AI across 8 markets

Led the language, grounding, evaluation, and rollout work that moved a 20M-user support experience from fixed rules toward measurable AI behavior.

25% faster responses60% fewer unsupported responses

Prompt architecture
Conversation systems

Read the case study

Knowledge and evaluation

Test what the system knows before customers do.

Case Study2024

Testing a support knowledge base before choosing RAG

Built a controlled evaluation around real support questions, expected citations, latency, and blind review before recommending a retrieval architecture.

Evidence for an architecture decision