Ben Whitfield

How I work — AI-native

• Claude Code is my daily, my primary development mode.
• Important context is preserved in a /ai_docs directory and updated constantly.
• Rapid Prototype
  • Spending time upfront to research with AI and clearly specify the requirements.
  • Claude researches existing issues with the mix of technologies we are planning to use.
  • Favorite prompt: Ask me questions until you are highly confident in what needs to be done.
  • Claude adds in deployment breaks and commits at checkpoints as part of the plan. We iterate and test in small increments to minimize complexity and keep debugging fast.
  • I challenge assumptions and ask what data is missing.
• Claude executes and we test as we go.
• I spot check the code and catch issues at our testing checkpoints.

Experience

I built Reazy, a cross-platform text-to-speech SaaS, from nothing to 500+ users, with paying subscribers across the Chrome Web Store, Google Play, and the Apple App Store. I was the only engineer — architecture, frontend, backend, ML training, payments, and deployment. The product turns documents, web pages, and raw text into natural speech.

Being the only engineer meant owning problems end-to-end:

• Frontend performance — built a virtualized reader that renders 1,000+ page documents smoothly, staying fast at any document size.
• Billing — unified Stripe (web), StoreKit 2 (iOS), and Google Play Billing (Android) into one provider-agnostic entitlement model: a single source of truth for who's paid, across all three stores.
• Machine learning, end to end — built the text-to-speech voice pipeline from data to production: data processing, models adapted from open-source architectures (FastPitch, HiFi-GAN), and a cost- and latency-optimized inference backend.
• Native mobile — built a custom native audio engine in Swift and Kotlin, replacing off-the-shelf plugins: gapless playback, lock-screen controls, and background audio on both platforms.

The app — web, Chrome extension, and mobile — shipped from one React + TypeScript monorepo with a shared core library and automated tests. Model inference ran as a separate Python service on Google Cloud Run.

Listen to any article, PDF, or document with natural AI voices. 60% less than competitors. reazy.pro ↗ Building Reazy: a cross-platform text-to-speech app How I built Reazy — voice models, cold starts on Cloud Run, native audio engines, a virtualized reader, and surviving a surprise Apple account termination. whitfield.dev ↗

Skills

AI / LLM

Generative AI · Anthropic API · agentic systems · prompt engineering · evals

ML

PyTorch · custom model training · production inference

Frontend

React · TypeScript · JavaScript

Backend

Node.js · Python/FastAPI · REST · Firestore (NoSQL) · Google Cloud Storage

Cloud

GCP · Firebase · Docker · Cloud Run · CI/CD

Mobile

Capacitor (iOS + Android) · custom native audio engine (Swift, Kotlin)

Testing

Playwright · Vitest

I am a dynamic and versatile developer. I will learn and conquer any skill needed for a project.

Background

Before software, I spent a decade as an operations-research analyst and economist for the U.S. Air Force, including at the Pentagon. I started teaching myself to build software in 2015 (Stanford's open courseware, The Odin Project) and went full-time in 2021: trying out mobile game development, a deep dive into web technology, followed by building and running Reazy end-to-end.