I build AI features that ship to real users — not demos. I've shipped a generative-AI app to the App Store and built the LLM pipelines and full-stack systems that power it.
A generative-AI photo editor live on the App Store, plus Trace's server-side Claude pipeline that turns a dropped pin into a written run. Led a 5-person team that lifted an LLM evaluation system from 50% to 96% accuracy — on an M.S. Mathematics (4.0) foundation.
Claude APIOpenAI APIPrompt engineeringLLM evaluationPyTorchCloudflare Workers
iOS Development
Shipped to the App Store, not just a demo
Sole engineer behind a published SwiftUI app: 3K users, zero crash rate, sub-6s cold start, on-device image processing, and an offline-first cache. Comfortable from concept to App Store Connect.
SwiftSwiftUIUIKitSwift ConcurrencyInstrumentsApp Store Connect
Full-Stack
From the UI down to the database
React/TypeScript frontends on .NET Core and PostgreSQL backends, serving 10K users in production. RBAC, REST APIs, and 85% test coverage.
Shipping products across applied AI, mobile, and the web. Filter by the kind of work you care about.
AIiOS
2025
AI Photo Editing
A generative-AI photo editor, shipped solo to the App Store.
An iOS app for editing photos with generative AI — built as the founding product of AI Editor LLC. I took it from concept to App Store publication as sole engineer in under three months. It now has 3K users and a zero crash rate across every production release.
A native macOS meeting assistant built on a real-time, on-device speech-to-LLM pipeline.
A native macOS app that listens to a live meeting and surfaces a single, well-timed suggestion rather than a running feed. I built the full real-time pipeline: system and microphone audio captured and transcribed on-device with WhisperKit, then streamed as rolling context to an LLM through a provider-agnostic client layer. The result renders in a floating overlay only the user can see. Built and shipped solo as an AI Editor LLC product; now in open beta.
Captures system and microphone audio and transcribes it on-device with WhisperKit through a custom batching pipeline with speaker-aware segmentation
Streams the rolling transcript as context and prompts the model to return one high-signal suggestion rather than a constant stream
Event-driven Swift 6 architecture (actors, async streams); floating overlay excluded from screen capture, with API keys stored in the macOS Keychain
Swift 6SwiftUIWhisperKitStreaming LLM inferenceMulti-provider LLM (BYOK)Prompt engineeringScreenCaptureKitSwift Concurrency
An AI-native travel app for runners — runs as destinations, not routes.
An app where every entry is a run with a story — a curated, Atlas Obscura-voice feed of iconic and hidden runs worldwide, plus a 'generate me a local run' feature: drop a pin, pick a distance and a vibe, and a Cloudflare Worker stitches a real walking loop through nearby POIs (Google Places + Directions) and has Claude write it up. I built it in two stacks: a native SwiftUI version and a cross-platform React Native (Expo) version running on iOS and Android — both implementing the same product spec.
Built in two frameworks from one product spec — native SwiftUI for iOS, and React Native (Expo) running natively across iOS and Android
Server-side generation pipeline keeps API keys off-device and produces identical results on every platform
Local-first state: bundled JSON for curated runs, AsyncStorage for saved + generated runs — no accounts, no server state
Led a 5-person team from 50% to 96% accuracy in Terence Tao's SAIR challenge.
Our entry in the SAIR Foundation Mathematics Distillation Challenge (organized by Terence Tao and Damek Davis): engineer a 1 KB plain-text instruction sheet that helps a small open model correctly decide equational implications over magmas, drawn from the 22M-pair Equational Theories Project. I led a 5-person team and built the evaluation harness driving it.
Led a 5-person team and built the evaluation methodology, metrics, and benchmark
Raised accuracy from 50% to 96% on a 300-problem benchmark, focusing on the hard subset where most teams plateau
Exploited structural priors (≈63% FALSE base rate, ~32% singleton laws); Stage 2 extends to machine-verifiable Lean 4 certificates
A full-stack web application for managing user groups, permissions, and custom attributes within an organization. RESTful .NET 8 backend, modern React + TypeScript frontend, PostgreSQL storage, and clean architecture with maintainable testing patterns.
Clean-architecture backend with a minimal, maintainable test suite
C#.NET 8ReactTypeScriptPostgreSQLREST
Private repo
Full-Stack
2026
P2P VPN
A peer-to-peer VPN where every user can also be an exit node.
A WireGuard-based VPN (an AI Editor LLC project) where users connect as clients or serve as exit nodes for others. A native Android client in Kotlin pairs with Go backend services; the tunnel implements the WireGuard handshake with dynamic key exchange and forwards both TCP and UDP traffic.
Exit node runs via BoringTun with no root required; client uses the WireGuard Android library
Verified end-to-end: HTTPS browsing and DNS resolution work through the peer tunnel
I'm an AI and iOS engineer who ships. As founder of AI Editor LLC, I designed, built, and published AI Photo Editing — a generative-AI photo editor — to the App Store as sole engineer in under three months, now 3K users with a zero crash rate across every production release. My AI work runs deeper than API calls: a server-side Claude pipeline in Trace that turns a dropped pin into a written run, and a 5-person team I led that lifted an LLM evaluation system from 50% to 96% accuracy. I also work across the stack — at AssetMark (financial services) I built end-to-end features in React/TypeScript, .NET Core, and PostgreSQL for a platform serving 10K users — all on a research-grade quantitative foundation: an M.S. in Mathematics (4.0 GPA).
Designed, built, and shipped AI Photo Editing (iOS) to the App Store as sole engineer in under three months — 3K users, zero crash rate, sub-6s cold start.
Built and shipped Insight (open beta), a native macOS meeting assistant: real-time on-device transcription (WhisperKit) streamed to a provider-agnostic LLM layer, surfaced in a screen-capture-excluded overlay.
Building a peer-to-peer VPN with a native Android (Kotlin) client and Go backend services.
Full-Stack Engineer
AssetMark · Concord, CA · Financial services
2025
Built end-to-end features across React/TypeScript, .NET Core (C#) with Entity Framework, and PostgreSQL for a platform serving 10K users, at 85% unit-test coverage.
Designed a scalable RBAC permission schema and shipped a legacy-integration layer (iframe + postMessage) bridging ASP.NET and new .NET Core/React apps.
Software Engineer
Practice Defenders · Remote · Electronic health records
2025
Built API integrations and automation pipelines across external EHR systems, focused on reliability and consistent data handling across distributed services.
Software Developer (Volunteer)
AESOP Afghanistan · Remote
2023 – 2024
Built Python backend services and API-driven workflows (Google APIs); refactored legacy systems and introduced automated testing.
Graduate Teaching Assistant
Northern Arizona University · Flagstaff, AZ · Mathematics
2024 – 2026
Taught and mentored undergraduate mathematics students, and mentored student teams in the DataFest data-science competition.
