I take consumer apps from zero to the App Store — full stack, payments included, AI tools development.
8 years of cloud engineering at Rocket Mortgage underpin the work. My daily loop is an AI coding agent, a test harness, and a release build — from design to the store.
// full stack, zero to store
A from-scratch agent harness in about 300 lines of Python, with no framework. It has a tool-calling loop, Model Context Protocol (MCP) tool servers added through a JSON entry instead of code, plain-markdown memory, and local or cloud models you can swap mid-conversation. Built to understand what tools like Claude Code and Cursor do under the hood.
Scores how well a property actually fits you. Buyers weight what they care about and enter budget, down payment, rate and location; fit recalculates live as the weights move. The system explains the reasoning — the buyer decides.
Upload a PDF, ask questions, get answers grounded in the document with sources cited. Ingest and query run as durable event-driven workflows, so long parses never block the UI, and deterministic chunk IDs make re-ingesting a file idempotent instead of duplicating it.
A containerized multi-agent service: one natural-language request routes through a LangGraph supervisor to agents that research a topic and send real email. Capability separation is structural — the research agent is never bound to the send tool — and user identity flows through run config the model can't see or forge. Live on DigitalOcean.
A manager dashboard for guest reviews with public property pages showing only approved ones. A channel-agnostic normalization layer keeps the UI from ever knowing a source rates out of 10 — adding a channel is one adapter, not a rewrite.
An AI "founding team" for early entrepreneurs: six mentors across twelve venture-building disciplines, delivered as Claude Code slash commands with a private, redaction-aware journal.
A Positive-Intelligence-based mental-fitness coaching stack for Claude Code: five AI coaches running a daily practice loop over a local, event-sourced journal.
Mobile apps — live on the App Store and Google Play
Every one of these went from idea to release build under my hands — architecture, code, payments, QA, and store review included.
A voice AI app that runs OpenAI's Whisper large-v3-turbo on the iPhone itself: no server, no account, no user data collected. I changed whisper.cpp to do half the work per audio chunk, making transcription real-time even on an iPhone 12 mini. Text appears as you speak, it catches language switches mid-sentence, and it keeps recording in the background with a Lock Screen display.
A peer-to-peer marketplace live in Mexico with 9,000 users and growing. Stripe escrow holds the buyer's payment until delivery; scanning a QR code at handoff verifies the sale and releases the seller's payout, with Twilio notifications along the way. Vision models turn a photo into a listing: title, description, category, and price label.
Turns live Instagram and TikTok data into a single 0–100 influence score — profile lookups, growth tracking, and shareable score cards. Free with in-app purchases.
An app that connects creators for instant collaborations. I built the agentic pipeline behind it: it scrapes and profiles creators and businesses across Yelp, their websites, and Instagram at up to 500 concurrent instances for about $0.04 each, with GPT vision writing descriptions, classifying categories, and screening images before a creator reaches a brand.
Founding Software Engineer
Technical Co-Founder / CTO
Software Engineer
The pipeline chains Yelp → the business's own site → Instagram, extracting contact and social data and pulling about two dozen images along the way.
AI does the judgment work: writing the description, classifying the category, filtering out national brands, and picking which photos are actually good enough to show.
Built at TRUST as technical co-founder. The video is a live demo, end to end.