Most of my work sits somewhere between native apps, backend systems, and AI tools.
If you only skim a few, start with Steward, Clanker, Scout, DigitalYou, NavRead, and Mac Use. The content pipeline is the odd one out, but it still helped a channel hit 35M+ views in two months.
Mostly macOS and iOS work: automation, operator tools, and products built close to the system.
A macOS automation app that records workflows and keeps them running even when the UI changes. It checks the screen a few different ways instead of breaking the first time a button moves.
usestewardai.com →A voice operator for Meta Ray-Ban glasses. You speak to the glasses, the request routes through the iPhone and an end-to-end encrypted cloud relay to your paired Mac, and a local runtime carries out the task. Managed model access and a signed private runtime mean no terminal, keys, or pasted secrets for the user.
A local-first macOS canvas for running isolated AI coding sessions side by side. Each session is sandboxed and talks to agents over the Agent Client Protocol, so you can spin up, watch, and compare runs without them stepping on each other.
An open-source macOS runtime that turns the Accessibility tree into something a model can actually work with, then exposes safe actions through a CLI, SDK, and MCP server.
Another take on the same broad problem as Steward: a local macOS operator that can read UI structure, use OCR when it has to, and act precisely.
useomnyxai.com →A desktop control panel for Codex-based Mac automation, with OAuth, saved profile data, local token storage, and a bridge back into the machine.
A native macOS subtitle app that listens locally and floats translated captions over whatever app, site, or video is on screen.
The iPhone companion app for Omnyx: auth, approvals, status, deep links, and push notifications. The App Store build is still up, but I pulled the API key and I'm not supporting it now.
App Store →An iPhone AR networking assistant for event floors. Point the camera at someone, say what you're after, and it locks the target, resolves identity, scores the lead, saves it to a memory graph, and drafts the follow-up — all from the AR lens.
Watch demo →Research tools, retrieval systems, and products with real backend work behind them.
A research agent for insurance brokers. It pulls public information, applies risk rules, and turns the whole thing into something useful before a call.
broker-scout.vercel.app →An AI communication platform for email, calls, SMS, and social. Memory, docs, integrations, analytics, testing. A lot of moving parts.
A private reading desk for PDFs, screenshots, OCR, notes, and retrieval. Basically a better way to keep research in one place.
GitHub repo →A local-first macOS PDF assistant. Open a PDF and ask questions — extraction, retrieval, and answering all run on-device with citations, and nothing leaves the machine. It answers from whole pages instead of stray snippets, and abstains when the evidence isn't there.
A local-first agreement review tool. Drop in a contract and it converts, renders, and segments the clauses onto an infinite canvas for reviewing highlights, relationships, and arguments. AI extraction is optional and validated against the source text before anything is saved.
A study-guide generator for teachers. Paste a Hindi, Hinglish, or English lecture link or raw notes, and it pulls the transcript, isolates the chapter, cleans and translates it, then renders an exam-ready PDF with Q&A, MCQs, tables, and a revision sheet — ready to hand to students.
A decision tool where AI helps set up the framework, but the user still controls the weighting and tradeoffs.
clearweight.vercel.app →A Google Business Profile management app for reviews, photos, AI suggestions, and multi-location reporting.
A conversational astrology product with memory on the frontend and a separate backend path for the domain-specific logic.
A retrieval app built around curated founder and operator voices, with pgvector search under the hood.
A crawl-and-retrieval pipeline that scrapes sites, embeds them, and makes the whole corpus searchable.
A prediction-market console with risk controls, analytics, and optional AI help layered on top.
A student-focused AI product with auth, pricing, payments, and task-oriented workflows.
An AI on-call engineer that resolves production incidents autonomously while holding zero standing credentials. Every destructive action needs a single-use, scoped, just-in-time grant, and the gate denies anything with too large a blast radius.
GitHub repo →A voice-driven search-and-rescue command center. LLM agents reason about who the missing person is, a Monte Carlo simulation computes where they likely are and how to search fastest, and a coordinator drives it hands-free by voice — re-planning as new information arrives.
Live demo →The stranger builds, where the interface mattered as much as the underlying system.
A video pipeline I built for a YouTube channel that reached 35M+ views in two months. Scripts, voice, animation, rendering, and throughput all had to work like a real production system.
A Chrome extension that lets you reshape live websites with natural-language prompts.
A side-project browser built to mess with tabs, UI, and privacy ideas that annoyed me in mainstream browsers.
A 3D comic-book version of this portfolio. Completely unnecessary. Also fun.
Read the comic →Not every project here has a public link. Some live in private repos, some came out of client or internal work, and a few are offline because I moved on and did not feel like pretending they were still active.
Last updated · July 22, 2026