Founder & builder | LK Integrate | AI context infrastructure | CMC '28
I build the tool first, then make the story catch up. Current focus: private context infrastructure for people who want AI systems to work with their real files without re-uploading everything.
Profile
Claremont McKenna '28
Economics + Financial Economics sequence, with a build-first portfolio across infra, fintech-style products, and consumer software.
Default motion
Idea to live product
The work includes auth, data models, gating, account state, downloads, signed-in flows, and public links.
Build stack
Next.js + Supabase + Vercel
Repeatedly used with OpenAI, Stripe, Electron, and MCP when the product needs payments, local context, or tool connectivity.
Selected work
Public links are included where they can be shared cleanly. The point is to make the work inspectable, not inflated.
AI context infrastructure
Live product
Built a full-stack product that syncs local folders, extracts and chunks files into a private index, surfaces grounded answers in a web dashboard, and exposes search + file retrieval through an MCP server.
Local file indexing, private cloud retrieval, dashboard chat, and MCP connectivity for ChatGPT-style workflows.
Prediction markets
Live product + public repo
Built a full market surface that combines discovery, account state, category filters, deposits, institution-only markets, and operational/admin layers in one product.
A prediction market product with public discovery, institution-gated participation, wallet flows, and resolution tooling.
Family memory product
Live product
Built a warmer consumer product around short voice prompts, guided life-story interviews, family circles, and searchable memories so preserving family history feels natural instead of like work.
A voice-first family memory app that helps families record stories, preserve milestones, and return to the people and moments that matter most.
Published tool
Reusable Codex skill
Public GitHub repo
A durable review workflow for AI coding agents. It saves the raw user request, the agent's best intent estimate, a change manifest, verification evidence, phased reviewer output, and machine-readable gate status so meaningful changes can be audited instead of trusted from chat memory.
Now
I am Michael Callow, a Claremont McKenna student building practical AI products in public. My strongest work sits where ambition has to survive contact with production: local file sync, private retrieval, dashboards, account state, payments, and agent workflows that leave an audit trail.
Writing queue
A public forcing function for a clearer LinkedIn presence: useful, specific, and tied to things that actually exist.
The strongest AI tools still lose context the moment your real files live outside the chat window.
Proof: Show the sync -> index -> MCP retrieval loop with a short product screenshot thread.
A student portfolio should prove shipping range, not just list clubs, classes, and buzzwords.
Proof: Use LK Integrate, There's No Chance, Remember, and Intent Verification Gate as the four anchors.
The hard part is not getting an agent to write code. It is proving it understood the request and verified the right behavior.
Proof: Walk through the Intent Verification Gate artifacts and why they exist.
Classes give constraints; production gives feedback. The combination is underrated if you actually ship.
Proof: Tie CMC, LK Integrate, and weekly launch loops into one honest founder note.