RadioPaint
A full-stack art-residency and LMS platform, architected solo — commerce, secure video, and a domain model rebuilt across several ground-up iterations.
The complete platform behind an online art-residency studio: a public storefront and gallery, real-money checkout against a Russian payment gateway, a secure S3-backed video LMS, and an admissions/entitlement engine modeled directly on how the business actually runs. Designed and built end-to-end — backend architecture, database, and frontend — working directly with the studio's founder as domain expert.
The Motivation
RadioPaint started life as a bare-bones DIY site builder bundled with T-Bank's own CMS platform — enough to take a payment, nothing close to what the studio's founder actually needed. I volunteered to architect and build a real platform for her, from scratch, to support where she wanted to take the business: live mentored residencies, a proper commerce and admissions engine, secure video hosting, and a portfolio system her own students could point a gallery or curator to. The project ran long enough that it went through several full rewrites as the business's ambitions grew alongside it — the fragmented, five-app state a later audit turned up wasn't inherited from anyone, it was my own earlier iterations. The current architecture is what came out of consolidating all of that into one deliberately-designed system: one domain model, one entitlement engine, one real payment flow — without losing any of the 300+ residents already using the platform, and without ever taking the site down to get there.
The Problem It Solves
By the time RadioPaint's business had fully found its own shape, the platform was still carrying the residue of the earlier iterations that got it there — revisions built, then superseded, over the course of a long-running solo build: SQL Server tables that no longer matched the C# models reading them, a React frontend built and fully mocked ahead of a backend contract that hadn't caught up yet, secure video with three stacked half-designs from successive attempts that were never fully wired together, and a checkout endpoint still computing its total from whatever price the client sent it. None of this was hypothetical cleanup — real applications, real event registrations, and real payments already existed on top of it, and had to survive being consolidated into one definitive design.