Founder Build • iOS • macOS • Web • Single‑Developer
I design and ship hybrid systems that make web tools feel native on iOS, Mac, and the browser.
My approach: patch what’s broken, intercept at the right layer, and add the missing primitives (offline queue, PDF workflows, secure auth).
I work across UIKit/SwiftUI, WKWebView, PDFKit, React, and glue code—end to end.
I build the other half too: React micro-UIs, macOS utilities, build tooling, and pipelines.
React/HTML: Micro-interfaces that slot into legacy pages without heavy frameworks.
macOS (SwiftUI/AppKit): Menu bar tools, PDF ops, file capture, quick-actions.
Pipelines: Preconnect & asset hygiene, cost-flags for AI vs deterministic paths.
Selected Demos
These clips highlight the internal app I built for my team.
What started as a way to wrap existing web tools on iOS evolved into a fully custom hybrid app:
a polished, stable environment that fixed framework-level issues on iPad and added entirely new
functionality.
From a branded weekly schedule builder to a built-in document editor with signing and export,
these demos show it’s more than just a wrapper — it’s a true app experience.
My current experience focuses on running smaller local models for experimentation and rapid prototyping.
These have been ideal for my hardware and allow me to quickly test and validate workflows.
For larger-scale environments or production workloads, I can adapt these approaches to API-driven or
hosted LLM solutions as needed.
Hands-on with small local LLMs (Ollama, Mistral, OpenHermes) for summarization, templating, and quick testing.
Design patterns and pipelines that can scale to enterprise-grade deployments when resources allow.
Modular architecture for easy swapping between local and cloud APIs.
Deterministic Systems — When AI Is Overkill
For some workflows, full-blown LLMs are unnecessary. I’ve designed deterministic systems using curated
phrase banks and rule-based transforms that produce fast, consistent, and “AI-like” results — without
the latency, cost, or unpredictability of machine learning.
Phrase banks that generate clean, varied, and cohesive outputs without relying on model inference.
Great for structured workflows like progress notes, standardized templates, or repeatable formats.
Feature-flagged logic to switch seamlessly between deterministic and LLM modes as needed.