AI Writing Tools at Scale

Embedded with writers, analysts, and product leads across teams to map existing writing workflows, identify friction points, and understand what 'good output' actually meant to each role. Shadowed sessions where possible; ran structured interviews where not.

Designed a review and approval layer that made AI suggestions feel trustworthy rather than opaque. Every AI output surfaced its source context and required explicit acceptance — no silent auto-fill that could erode accountability.

Collaborated directly with researchers from the MIT AI Accelerator to align UX patterns with model behavior. Translated technical constraints (latency, confidence thresholds, hallucination mitigations) into interface decisions users could actually reason about.

Built the prompt-side experience: guided input patterns, context injection controls, and side-by-side draft comparison. Paired with an internal eval dashboard so teams could measure output quality and flag regressions.

Led adoption across 16+ product teams with role-specific onboarding materials and feedback loops built directly into the product. Iterated rapidly on pain points surfaced in the first 60 days post-launch.