I led the design and research of A to Z Assistant (Amazon Q) a generative-AI assistant that streamlined work for both AWS enterprise customers and Amazonians. Before Amazon Q, people wasted time switching between countless tools and data sources to complete a single task.
We created a single co-pilot experience that brought those systems together - surfacing insights, automating workflows, and enabling users to create lightweight task apps with natural language. Amazon Q integrated securely with internal Amazon tools and AWS services, delivered cited answers, and allowed actions to be taken directly from the chat experience or embedded surfaces like Slack and internal dashboards.
6 months
Time
6
User Tests
20
Plugins Integrated
+30%
NL queries
70
CSAT
my role
- facilitated stakeholder reviews with S-Team (Andy Jassy) and AWS SVPs
- led ux vison and delievrables for the end-to-end co pilot experience across platforms
- ran cross-org research to define use cases and AI intergation framework
- co-authored extensive LLM literature review which supported other teams across orgs.
- collaborated with ML/infra teams to shape plugin and RAG experiences
- authored generative AI design tenets and interaction principles
- helped define and test the MVP plugin framework for pilot launch
how it works
context aware help
adapts to users org, role, and location — no dead ends, just smarter help.
built to scale (agents)
users could complete tasks like booking PTO, generating docs, or opening IT tickets all inside the chat window.
shows up where you work
integrated across a to z, slack, and internal apps using a plugin model, letting teams build and own domain-specific skills while keeping the UX unified.
answers you can trust
combined amazon q’s enterprise rag with curated content sources to provide accurate, sourced, and safe responses.
transparent ai
clear disclaimers, transparent sources, human fallback options, and continuous feedback loops keep AI safe and reliable.
AI-powered assistant providing contextual help across Amazon tools
natural language AI conversations
intelligent contextual suggestions
educational affordances for agent visibility
converstional patterns for intention detection agent ux
overview of integration strategy across core surfaces
what I learned
transparency and education builds trust
people won’t rely on ai until they understand what it can and can’t do.
context is king
generic answers felt dumb, contextual ones were useful.
conversation alone isn’t enough
memory and proactive actions were the differentiator.
ai works best when it blends in
people didn’t want a new tool, they wanted something that shared the same context, to speed up their work.
designing ai requires new types of collaboration
working closely with infra and ml teams made the ux actually work.
redefining what success looks like
we had to find new ways to track “time saved”, not just clicks and usage.
overview of integration strategy across core surfaces