Katie O'Connor

Summary

Senior software engineer and design-systems architect. 11+ years at Discovery Education on products reaching 45M students and 4.5M educators across 100 countries, architecting three successive generations of the company's component library. Now building AI-native applications end-to-end, from local vector-search retrieval systems to full 3D game architecture, pairing deep design-systems and accessibility expertise with daily AI-assisted engineering.

Skills

AI Engineering: Prompt Engineering, Spec-Driven Development, LLM Cost/Model Strategy, Prompt Caching, RAG, Vector Databases, Anthropic Claude API, Claude Design
Frontend & Architecture: Vue 3, Nuxt, TypeScript, Design Systems, Component Architecture, Theming Architecture, Tailwind CSS, Vite, Accessibility (WCAG 2.2, ARIA)
Backend & Infrastructure: Python, FastAPI, ChromaDB, Node.js, Docker
Other: User Experience Patterns, Cross-Team Technical Leadership, Figma, Git, GitHub, VS Code, Cloudflare

Experience

Senior Software Engineer

2015–2026

Discovery Education · Chicago, IL

Design Systems

Architected nebula-nuxt, a themed component system letting teams replace legacy UI incrementally while migrating to Nuxt, separating visual-language changes from component rebuilds; selected over competing proposals as the quarter's committed deliverable.
Led the migration audit and stakeholder roadmap for 46 legacy component replacements; personally built 25 of the 26 components delivered in the new system and released nebula-nuxt for platform-layer consumption.
Founding team for Comet (2016), build team for its successor Nebula (2020), and owner and primary contributor for nebula-nuxt (2026); managed 124 components and utilities across Nebula and nebula-nuxt simultaneously, consumed by every product in Discovery Education's app ecosystem.
Built the nebula-nuxt doc site with searchable documentation (Nuxt Content) and a theme editor that switches themes and customizes them down to semantic tokens and font choice.
Architected and shipped MegaMenu for a Back-to-School product deliverable in 3 days, then delivered a refactored production version in another 3; detects at runtime when to collapse into a linked drill-down list rather than relying on fixed breakpoints.
Developed custom Claude skills to support migration: a planning skill that determines the right build approach (themed primitive, wrapped primitive, or fully custom component) for each legacy component, and a release-automation skill.

Product Engineering

Contributed as a product engineer to 12+ product teams across the app ecosystem, including Techbook, Search, Assessments, Admin Tools, and User Onboarding; architected layouts, interfaces, and components across varied product domains.
Embedded with the AI team for a quarter; built the frontend for a human-in-the-loop dashboard for reviewing and approving AI-generated metadata, completing the partial design and presenting it to real users for feedback.
Specified the data shape for the API feeding the dashboard view and the POST contract for saving reviews, documenting both for the engineer who built the API.
Integrated nebula-nuxt into the platform layer with ten production components, including Global Nav, Page Shell, and Modal; validated theming through a live-switcher playground.
Designed production-ready interfaces directly in code with Claude and existing design-system components when designs were incomplete or unavailable, bridging design and engineering to keep product work moving.
Advised engineers on theming architecture, component placement, and accessibility best practices; redirected in-progress work into the shared library.
Served as accessibility subject matter expert, performing audits, bug fixes, and code review to enforce accessible implementation across products and shared components.

AI-Assisted Engineering

Automated end-to-end component delivery (build, doc page, and release) through spec-driven development and skill-based automation; $3.05 in AI spend per component, replacing roughly a day or two of hands-on engineering.
Developed model-selection and prompt-caching strategies that reduced daily AI spend by roughly two-thirds, from $40–$50 to $10–$20, while preserving output quality; shared the cost-efficiency approach across the team.
Co-hosted the company's AI Office Hours, and designed and presented a cost-efficient stacked-PR strategy there, structuring PRs in advance to reduce rebasing.

Projects

Main Character: AI Personal Knowledge & Retrieval System

Python, FastAPI, ChromaDB, Anthropic Claude API

Architected an AI-powered journal application using local embeddings, an entity graph, multi-level summaries, and semantic retrieval over personal journal entries.
Built the retrieval pipeline to ground LLM responses in the user's actual journal history rather than generic model knowledge.

GPS3 (Guinea Pig Simulator 3)

Vue 3 (Composition API), TypeScript, Pinia, Three.js, Vite

Solo-developed virtual pet care simulator managing 10 interconnected wellness needs across 2D game view and interactive 3D habitat.
Designed game's visual language in Claude Design, then converted into a standalone token-based design system and ported back into the game as the single source of styling truth.

Earlier Experience

Front-end & Production Roles

Chicago, IL

Sears Holdings Corporation, Razorfish, AlphaZeta Interactive, Elevation, Hewitt Associates, Bankers Life and Casualty, iPort Media, WGN-TV.

Education

BA, Interactive Art and Media

Columbia College Chicago