ConnectMusic is a music publisher. Deal360, its internal app, values song catalogs and runs publishing deals from first contact to underwriting sign-off, across three roles, including sales and the underwriter.
It’s an internal tool, so there’s no public link. The data in every shot is made up.
Mara Vey — Back CatalogSend to underwriting
Mara Vey — Back CatalogShares reviewArtistCreated by Owen HaleSep 28, 2026Week 42
Songs in deal54 conflicts15/24 of Luminate catalogIn conflict first1FiltersColumnsModify songs
Lantern Song (Live)1,341,2802,118,65030,420−10%25%50%50%50%
All songsIncompleteCompleteIn conflict
Claims overlap the share: 75% + 50% = 125%
Biggest challenge
Tight timeline, small team and three roles to cover, each with different needs. On top of that, the app had to replace a lot of third-party tools in one solution, to take away the friction of switching context between YouTube Studio, Luminate (goated tool to review music data in-depth) and the others.
Syncing from Luminate03691215182124 of 24 tracks · ~1 minSync complete24 tracks loaded15 tracks selected
New dealDeal typeArtistWhole catalogProducerSong by songArtist catalogFrom LuminateSearch artistMMaMarMaraMara VeyAR-3F9C07ClearArtistLuminate IDMara VeyAR-3F9C07Mara Vey & The LanternsAR-81D4A2MaraluneAR-C05E8BYoung MarabelAR-6A2F0DRequested advance$05505005,00050,000500,000USDCancelContinue
How I’ve used AI here
We started with a pretty basic process in Figma to set the direction and the high-level UX patterns, plus a few (many :P) workshops to map the needs of each role.
Once the project had a proper set-up in the codebase, I dropped Figma completely. I prepared screens in Claude Design and later in Claude Code, and delivered almost ready-to-merge code for each feature. Our front-end developer reviewed it and checked it against the API contracts, so we could connect it to the backend.
It was also the first project where I worked this closely with a back-end developer. I had to understand the limitations and the way they work to design things that were doable. I ran a Docker image of the backend on a seeded database for the first time, too.
AI wrote the prototype code. The layout, the interactions, the copy and which states exist were my calls. The longer story is on How I work, and still no AI slop, I promise.