Product
Engineer

Foreclosure Radar tracks every property heading to a forced sale in Pennsylvania and New York. Sheriff sales, county tax claim upset sales, county tax auctions.

Whoever bids on one of those houses is deciding almost blind. A list of the sales is free everywhere, and it won't tell you what the place is worth, what shape it's in, who still lives there, or which debts stay with the house after the gavel drops.

So instead of printing another list, the site does the homework on each lot. Every lot with a photo gets a written read on its condition. I fix the Pennsylvania numbers before anyone sees them, because one county out there still values houses on a 1961 base year, so its own figure means almost nothing.

The flipper I build it with said the line that changed how I think about the whole thing: "I am going to buy this house, but how much am I going to sell it?"

It also won't print what it can't prove. A court record only belongs to a property when an address matches, so on one sale list the search came back with 1,379 name matches and I kept none of them.

It covers 3,812 properties across 48 auctions right now. One design partner, nobody paying yet. I built the payment side and left it switched off on purpose. It is not open to the public while that is still true.

Job hunting is mostly retyping. Same name, same phone number, same eight screener questions, on a slightly different form every time.

The auto-apply engine behind Jobbi does that part for you. You see a job, you press Apply, you see what happened. Underneath, it opens the employer's real application form, fills in the identity and legal fields from what you told it once, answers the written questions using only your own resume, attaches a resume written for that posting, and sends it.

I pointed it at my own job search first. It filed 1,183 applications for me, end to end, 52 waves over 18 days. A month of model calls came to about three dollars.

Most of the design is about what the model isn't allowed to do. It can't invent a degree or a job you never had, because it only ever sees what's in your own resume. Equal opportunity questions never reach a model at all.

The best thing I've built into it is a refusal. For a long stretch it only really worked for me, because mine was the only complete profile in the system. Now it checks first and tells you what's missing in a couple of seconds, before it opens a browser and before it charges you anything. Not writing that check sooner cost me a month.

Upload a resume. Paste a job link. Get a version written for that job in seconds. No prompting, no setup, no settings to learn. That was the whole idea, and it is still the first thing the product does.

It has grown a long way past that. Jobbi now scores how well a resume matches a posting and names the words it is missing, writes the cover letter, lets you set your own rules for how it writes, and gives you the same resume at three lengths depending on who is reading. It runs on the web and on iPhone, and the two stay in sync. 1,537 registered users and paying subscribers, all from organic traffic, nothing spent on ads.

Then it started applying for you. The auto-apply engine opens the employer's real form, fills it from your profile, answers the written questions from your own resume, attaches a resume written for that posting and submits. That part has its own entry above.

Privacy is in the architecture, not in the policy page. Personal details are stripped out before anything reaches a model and put back in your browser afterwards. Your name and address never touch a third party's API. People told me that is why they picked it.

Built every layer myself: iOS in SwiftUI with SwiftData, web in Next.js 16 and React 19, backend on Supabase and Deno, three model providers behind one service layer so the right one runs each task and you never see the switch. Apple Vision reads a photographed resume, PDFKit parses it, RevenueCat keeps the subscription the same on the App Store and on the web.

Seven tools for job seekers, six of them AI-powered, live in production. ATS Resume Scanner. Resume Roaster. LinkedIn Roaster with shareable cringe-score results and dynamic OG images. Resignation Letter Generator. Invoice Generator. Markdown-to-PDF with live preview. Word and Character Counter with a live meter for every platform limit, all of it client side.

Built on Next.js 16, React 19, and 30+ Radix UI components. Supabase for auth and storage. Deno Edge Functions for AI processing. Dynamic OG image generation on the Edge runtime. Blog with admin panel. Resend for email. End-to-end TypeScript across client and server.

RecipeShare is a React + TypeScript single-page application built with Vite, Tailwind CSS. It uses Supabase for authentication, database, and serverless edge functions powering features like Stripe Checkout, Stripe Connect onboarding, content moderation, bulk user/recipe imports, and admin user export. The app includes a full marketplace flow with recipe purchasing, seller dashboards, referral tracking, user analytics, and profile management (favorites, saved, purchased, and following). State is organized via custom hooks and a centralized AuthContext, with modular services handling recipes, profiles, analytics, and interactions. The codebase is production-oriented with SQL migrations, PostCSS, and a clean page/component architecture.

Built Prompt Crafter Club, a platform to create, organize, and publish structured AI prompts with pricing, metadata, and versioning. Backed by Supabase for authentication and Postgres data (RLS policies). Stripe powers checkout/subscriptions and webhooks for paid access. OpenAI integration enables prompt testing and preview generation. Google API integrations support OAuth and import/export. Includes a guided editor, creator profiles, and bulk JSON import/export to seed or manage libraries. Implemented strongly typed React + TypeScript forms with validation and reusable UI components - focused on maintainability, clarity, and scalable state flows.

An innovative time tracking and invoicing app that enables real-time tracking, automatic earning calculations, and detailed reports for freelancers and small businesses.

Rebuilt the real-time tracking map with Azure Maps: intelligent clustering that groups live points by status, breaks into smaller clusters on zoom, customizable pointers, labels, a screenshot generator, and polygon area selection. Live status updates over SignalR. Paired with an operations dashboard on AG-Grid Enterprise: custom renderers, custom actions, infinite scroll with virtualization, and persistent per-user settings. TDD throughout. Internal enterprise software, so no public gallery.

Led the frontend migration from Material Design (MUI) to Tailwind CSS: smaller bundles, consistent design tokens. Built a Shimmer Loader component adopted by product teams across the organization, and contributed to the company-wide Angular UI component library built on signals. Internal enterprise software, so no public gallery.

Took over a legacy AngularJS application customers paid for: stabilized it, fixed production issues, and earned consistently strong customer satisfaction scores. Then drove the migration to modern Angular (signals) and React micro front-ends over GraphQL, backed by Node.js BFF services. Established the engineering quality foundation: Husky pre-commit hooks and GitHub Actions CI/CD with mandatory unit tests. Internal enterprise software, so no public gallery.

2-in-1 smart reminder and Bluetooth device finder with notifications, free device finder with ads, and simple one-time purchase for reminders.

iOS app using Texas DPS open API to find nearby appointment slots; NestJS backend for scheduled checks and push notifications.

UI for an AI-driven system that reads license plates and links them to live/recorded video footage for monitoring and security operations.

Designed and developed the user interface for an AI-powered search tool allowing users to locate individuals by entering physical descriptors, uploading photos, or selecting examples from recorded video. Improved user experience for security and investigation workflows by integrating advanced visual search capabilities into surveillance systems.

Developed the user interface for a real-time AI-powered crowd detection system to monitor and alert on unusual gatherings or when crowd size exceeded set thresholds during the COVID-19 pandemic. Improved user interaction and contributed to public safety by integrating the UI with existing monitoring infrastructure.

Enterprise app integrating complex APIs and workflows; led decoupling to microservices (.NET Core backend) and React frontend for scalability and performance.

NADA demo app with Angular Material UI and Symfony backend using Abstract Factory pattern for flexible APIs; responsive, modern interface.

Multi-language web app with custom UI components; backend in Symfony handling email processing and XML-to-JSON conversion.

Custom CMS with Angular + Bootstrap 4 and Laravel API secured with JWT; modern responsive UI.

Custom plugin to exclude internal IP ranges from Google Analytics, improving data accuracy.

WP/WooCommerce plugin adding Google reCAPTCHA for secure login; integrated with existing auth flow.

Plugin injecting user's ISP details into emails based on IP lookup for tracking and messaging.

Internal CMS for managing video content; Laravel backend with AD integration, Bootstrap + jQuery frontend.

Highly customized WooCommerce site with complex variant selection and automated email marketing plugin.

Company presentation website on WordPress; front‑end UX improvements and backend integrations.

Internal website with Active Directory sync via custom cron-driven integration to keep employee data up to date.

Internship project focusing on UI optimization and smoother app performance; contributed on both front‑ and back‑end.