Max
Kucera

A desktop tool for a local nonprofit that was keying bills by hand. A local Flask app sends each PDF or photo invoice to the Claude API for structured extraction (vendor, date, line items), fuzzy-matches the vendor against their canonical list, and, after an on-screen approve / edit / reject review, writes only approved rows into their Excel workbook via openpyxl, with automatic backups and an audit log. Claude only reads the invoices; the app does all the writing.

A browser-based Milkdrop visualizer running the real projectM engine compiled to WebAssembly, rendered through WebGL2 and driven by live FFT from the Web Audio API, off an uploaded track or the mic. Built in React, TypeScript, and Vite; it exports video via WebCodecs and ships an MCP server so AI agents can generate and load .milk presets into a running session.

An iOS app (React Native + Expo) for AI vision boarding: upload a selfie and describe a goal, and a Supabase edge function calls Google's Gemini image model to render your likeness inside that realized scene, then generates affirmation mantras with spoken audio. Backed by Supabase for Postgres, storage, and auth, with subscriptions through RevenueCat.

A restaurant-ops web app (React, TypeScript, Supabase). Operators upload vendor invoices; a Gemini-powered edge function OCRs them into line-item costs that feed fuzzy-matched inventory, recipe costing, and margin-based menu-price suggestions. Stripe handles subscriptions.