Staff-level engineer focused on AI developer experience and local-first tooling. I build the things that help developers enter flow state and stay there.
Founding member of the AI Advance Team — a cross-disciplinary group helping engineering teams across Disney adopt practical AI workflows. The projects below trace this work.
Senior Software Engineer on the Hulu and Disney+ Web product surfaces. TypeScript, React, Next.js. Shipping the core browsing experience on hulu.com — most recently the Video in Browse tiles and the styling-layer migration from SASS to vanilla-extract.
A native macOS app that turns your ChatGPT data export into a searchable archive. Hybrid search combines FTS5 keyword matching with on-device semantic search, fused via reciprocal rank — find conversations by meaning, not just words. Embeddings are generated by llama.cpp and stored in sqlite-vec.
Conversations render as clean Markdown with inline images, voice transcripts woven into the dialogue, and links to the original audio files. An "On This Day" feature resurfaces what you were thinking a year ago. Fully offline after a one-time model download. App Sandbox enabled. Built with AI, ships AI.
A local-first developer tool implementing the Model Context Protocol for Disney's engineering workflow. 48 tools across 10 categories: semantic code search via ChromaDB and CodeBERT embeddings, AI edit change tracking, git workflow integration, npm monorepo support, and full web project analysis. All running on the developer's machine, not in the cloud. Designed for simple npm/git-clone distribution across Disney engineering teams.
A native macOS application for managing, editing, and sending prompts to AI assistants. Built with SwiftUI and ProseMirror, Promptdown treats prompts as first-class documents — taggable, searchable, and reusable — with direct integration into Claude Desktop and ChatGPT via accessibility APIs.
Created and hosted an internal podcast at Disney covering MCP server development, agentic runbooks, and the prompt-engineering patterns I rely on day-to-day. Two seasons, nine episodes and counting — built to shift engineering culture, not just ship code.
A speech-to-text tool I shipped with AI as my engineering partner — I didn't write Swift; the models did (mainly Claude Opus 4.6). I brought the spec and the judgment calls. Built on whisper.cpp, runs entirely on-device. Reads the focused app's accessibility tree at dictation time and prompts whisper with what's on screen — so code identifiers, file paths, and terminal output transcribe correctly. Used by Disney engineers for voice-driven coding, prompt writing, ticket drafting, and code review notes.
A comprehensive architecture specification (v8) for a human-in-the-loop AI orchestration system. Coordinates agents, artifacts, and decisions through filesystem I/O and append-only event sourcing. Restart-safe and crash-recoverable by design — workflows survive agent failures, machine reboots, and session loss, then rehydrate from canonical artifacts and event receipts. Features a monitor agent for external health observation (because derailed agents can't diagnose themselves), autonomous correction before human escalation, and Git worktree isolation for concurrent workflows.
I've been creating software since 1996. I started at Microsoft as an Interactive Media Designer doing motion design, moved into Program Management, and shipped Windows Movie Maker v1.
I left after exactly four years to the day — I wanted more innovation than I was finding there. Two startups followed, back-to-back, as a Program Manager: Action Engine (8th employee) and SNAPin (2nd employee). I'm a co-inventor on multiple US patents for mobile customer self-support methods, originally filed via SNAPin and now held by Microsoft Technology Licensing. From there, three and a half years as Director of the Platforms & Technologies Division at Smashing Ideas, where I also wrote a lot of code.
Then I went all in on Spiral9. Using proceeds from the SNAPin sale, I built it into a full-service shop — ideation, design, development, deployment — staffed with designers, developers, and program managers across the lifecycle. We scaled to 14 people for about three years. Clients included Adobe, TiVo, Cisco, Facebook, and Disney. I pulled it back to solo work after that and have really been enjoying being an IC staff engineer.
The decades since 1996 have spanned a lot: web, mobile, desktop, connected TV, backends and frontends. I've collaborated making software from three vantage points — designer, program manager, developer — and I've found all of that experience super useful while collaborating with AI agents to make software for the last 3+ years.
AI is the most exciting technological advancement in my lifetime, and I'm really enjoying being part of it.
Now I'm focused on joining an AI company — building AI products and helping shape how humans and AI collaborate.
Open to conversations about AI developer experience and product engineering roles.
keldon@spiral9.com