ClearPing

What it is

ClearPing does one thing and attempts to do that well: takes an image with an unwanted background, removes it, and hands back a clean transparent PNG. From there it has a full workflow to edit the result - so you're not stuck with a one-click output you can't refine. It's deliberately narrow. It is not a full design tool, and it isn't trying to become one.

Access is invite-only with a magic link sign in, which keeps the web app clean and private by design. There's a separate admin surface for managing users and invites, with operational visibility over how the service is running and full usage statistics.

Why it exists

This project was a proof of concept to see if I could do full product delivery using agents. But not through typical personal vibe coding; this implemented a full dev cycle similar to what I've managed at work. Additionally I wanted to prove out my ability to create a full web app solely on my own. If I could take a specific product idea and build it, then I would have a blueprint for future projects with learnings.

The product

Release 1 was a narrow internal proof of concept: automated background removal, preview, download, strict temporary-file cleanup, and a small admin surface for access control and operational visibility.

Release 2 was the first real-user-testing release. It adds an editing layer - remove or restore specific regions, with a responsive editor that changes layout by screen size. Desktop uses a focused zoom canvas; mobile uses a bottom-sheet layout. Every edit request is validated locally before it hits the worker, and each selected area is saved as its own edit with a final "send all edits" step. The goal is a light, focused experience, not a broad photo editor.

Release 3 was a UX overhaul for the image editing journey. This release hit a product blocker - we were sacrificing a good user experience to try to work around image editing libraries that weren't giving us the functionality we needed. We spent more time fighting the tech than moving the product forward. Our real problem here is I didn't want to pour a lot of money into this project, and the access costs for API's for more powerful image editing just isn't warranted. This release hasn't yet made it to production.

How it's built / the stack

The stack was chosen for fast execution, quality, and low operational overhead: React + TypeScript + Vite frontend with separate user and admin entry points, Supabase for magic-link auth, Postgres, and minimal temporary object storage, Vercel for app hosting, and Oracle for API orchestration with a provider adapter layer for the background-removal engine. All code is stored in GitHub with automated deployments to Vercel and Oracle.

The product policy is private by default: no retained file storage, temporary assets deleted aggressively after download, untouched assets cleaned up after 20 minutes, short-lived signed URLs, and no raw images in logs. Accept JPG, PNG, and WEBP under 10 MB.

Full product requirements and technical design were written in Notion first, then built with agents handling the heavy lifting. Mockups were provided by agents for each screen before building. We tracked each task on a board and had one orchestrator agent defining architecture and handing off coding tasks to two other agents. Automated testing first ran via CLI, then via browser automation through Playwright, followed by a manual UAT cycle.