Emery Reszka

1. 1

   WingIt!

   An AI texting assistant for web, iOS, and Android, built from one codebase.

   You upload a screenshot of a conversation. It works out who said what and drafts a reply in the other person's tone. A coach answers questions about that thread, and a mode called The Read describes how each side is coming across.

   The model's wrong answers sound as confident as its right ones, so most of the work went into prompts: tightening them until replies came back in a format the interface could depend on, and deciding what to show when they didn't. Screenshots are deleted after processing, so there is no saved set of conversations to test a prompt change against.

   • React Native
   • Expo
   • TypeScript
   • Python
   • Claude API
   • Prompt engineering
   • Fly.io
   • REST APIs

   Open the live app

   

   Status: Live

2. 2

   Pulse

   A dashboard that tracks several unrelated markets at once. Live since January 2026.

   It puts stock prices, Polymarket prediction markets, and sports betting lines on one screen and flags a market that moves when the other sources don't.

   Most of the work was on the data. Sources disagree with each other, fields disappear without notice, and rate limits cut polls off partway through. I still can't always tell a real spike from a bad row that got through.

   • Python
   • Multi-source aggregation
   • REST APIs
   • JSON
   • Vercel

   Open the live dashboard

   

   Status: LiveJan 2026

3. 3

   RegionOS

   A Windows tool that shows live previews of any region of your screen. Version 0.1.0 came out in July 2026.

   You draw a box around part of the screen, an application window, or a website, and the preview keeps updating, even while the source is covered, minimized, or moved off-screen. Fullscreen mode hides everything except the grid of previews.

   Windows doesn't give you an image of a window it isn't currently drawing, and most of the code exists to work around that. Recognition runs offline, using deterministic computer vision instead of a model. It recognizes less than a model would, but it fails the same way every time, so I can reproduce a bad result and debug it.

   • Python
   • Win32 API
   • Computer vision
   • PyInstaller
   • Windows desktop

   Read the source Download v0.1.0 for Windows

   

   Status: v0.1.0Jul 2026

4. 4

   SnipAI

   A Chrome extension that re-reads part of a browser tab whenever it changes.

   You select a region of a tab, and it sends that region to a vision model each time the pixels change. You can switch between Claude, GPT, and Gemini while it runs.

   Manifest V3 replaced the persistent background page with a service worker that Chrome can shut down at any time. Watching the region, detecting a change, and calling the model all had to recover from being restarted partway through.

   • JavaScript
   • Chrome extension
   • Manifest V3
   • Vision models
   • REST APIs

   Install from the Chrome Web Store Read the source

   

   Status: Web Store

5. 5

   Campus event management system

   A Django application built by a team on sprint cycles, where I owned registration and waitlists.

   Accounts and roles, event registration, waitlists, and attendance tracking, built for one organization rather than the public. Frontend and backend were split between separate groups.

   Waitlists took longer than I expected. Storing each person's position as a number makes reads cheap, but one cancellation in the middle means updating every row after it. Role-based access touched the same rows: an organizer and a student looking at the same registration are allowed to see different fields.

   • Django
   • Python
   • SQL
   • Authentication
   • Role-based access
   • Agile sprints

   Read the source

   Status: Source

I've also made a turn-based JRPG in GameMaker Studio 2 and a top-down RPG in C++ with SDL2, both mainly to practice procedural generation. They're in the basement, at the bottom of this page.