Projects

Problem

Accessibility expertise is often trapped with specialists, while developers, designers, and product owners need contextual guidance at the moment they are making decisions.

My role

Conceived and built the product direction and AI workflows as Lead AI Engineer at Deque.

Approach

Used retrieval and workflow design to adapt authoritative accessibility knowledge to the user's task, role, and work surface rather than giving everyone the same generic answer.

Proof

Brought accessibility guidance into places teams already work, including Slack, VS Code, web, and Microsoft Teams, without pretending AI replaces expert judgment.

Problem

Higher education teams need to make large bodies of course material usable without treating every supported document as a slow, one-off manual remediation project.

My role

Founded and led the Enablement Engineering build across product direction, architecture, prompt and evaluation design, and stakeholder delivery.

Approach

Designed a semantic transformation workflow: document extraction, narrow AI procedures, human review checkpoints, PII safeguards, known-limitation boundaries, and auditable outputs.

Proof

For supported document types, the workflow reduced remediation cost from roughly $9/page to about $0.09/page while keeping human review in the loop.

Problem

AI systems are only as useful as their context, but useful source material lives across websites, PDFs, documents, media files, and browser sessions.

My role

Designed and built the conversion model and practical workflow for making source material reusable, readable, and reviewable by AI systems.

Approach

Normalize heterogeneous sources into structured Markdown so agents can work from inspectable text instead of opaque files and scattered tabs.

Proof

Makes context gathering a repeatable, reviewable procedure rather than a hidden pile of pasted files.