Every hobby has a part you love and a part that makes you wonder how to make it faster.
For me, DJing has both. The part I love is the performance: reading a room, mixing energy, getting a crowd to feel something. The other part is prep. Before a gig I have to pick the 15 or so tracks that fit a house party or a club opener, check that the BPMs build naturally, keep the keys compatible and plan an energy arc that peaks at the right moment. That prep is where I saw room for AI.
So I built SetLab.
What Is SetLab?
SetLab is a full-stack AI copilot for DJs. You describe the gig (genre, mood, crowd, set length and lineup slot) and SetLab builds a tracklist from your own library in seconds. Every track comes with BPM and Camelot key data, and the set is drawn as an energy arc so you can see its shape before you load a single track.
The name is literal: “Set” as in a DJ set, “Lab” as in a place to experiment.
Unlike a playlist generator, SetLab exports straight to Serato crates or Rekordbox XML and M3U, the formats the two main professional DJ apps use, so the set drops into the software you already play with.
Why I Built It
I’ve DJed as DJ Kelton Banks for years. DJ hardware, software and music libraries have all improved in that time, but set planning still happens by hand: digging through crates, checking BPMs, cross-referencing Camelot keys, sketching an energy curve and hoping it works for the room.
The work isn’t hard, but it’s slow. Planning a 60-minute set for a specific crowd can take 30 to 45 minutes, and that adds up across regular gigs and residencies.
I put SetLab through the test I use for any product decision: does it solve a real problem, or am I building for the sake of building? It solves one. The output is concrete, and because it exports to the tools DJs already use, you can play it that night.
How It Works
SetLab runs a three-stage pipeline on Claude, and it only picks from tracks in your library.
Here’s the flow:
- You set the brief. Mix name, primary and secondary genre, optional mood, crowd (club, lounge, wedding, festival, house party, radio or corporate), set length (30, 60, 90 or 120 minutes) and lineup slot (opener, middle, headliner or closing). You can add a venue for context.
- SetLab builds the set. Plain code filters your library first: genre and era rules, a clean-only option and a readiness check tell you up front when your library can’t fill the set. Claude Sonnet 5 then plans the energy arc and BPM ranges, and picks and orders tracks from the filtered pool. A reviewer pass checks transitions and key compatibility, and Claude Haiku 4.5 writes a note on each track explaining why it’s there and how to mix into the next one.
- You review and export. Regenerate or adjust the set, then export it to Serato crates or Rekordbox XML and M3U.
Outside set building, SetLab has an Explore section for track discovery by BPM range and genre; Track ID, which uses your browser mic and ACRCloud to identify a track playing in the room; library analysis that uses live web search to find emerging artists in your genres; and a library and wishlist for managing crates over time.
How it’s built
SetLab.ai: from a gig brief to a playable set
A three-stage Claude pipeline that only picks from tracks you own, with plain code handling the music theory checks.
- Input
- AI step
- Human checkpoint
- Code
- Import your libraryInput
Your Serato or Rekordbox library is parsed and saved, then enriched with genre tags, BPM and key.
- Serato
- Rekordbox
- Last.fm
- Supabase
- Describe the gigInput
Genre, era, crowd, duration and lineup slot. Optional seed artists or a source playlist.
- Next.js
- Filter the poolCode
Genre and era gates, a clean-only filter and a readiness check stop a set the library can't fill before any AI runs.
- Camelot keys
- Genre rules
- Blueprint the setAI step
Claude lays out the energy arc, BPM ranges and phases for the gig, returned as structured data.
- Claude Sonnet 5
- Forced tool use
- Select and sequenceAI step
Claude picks and orders tracks from the filtered pool, then a reviewer pass checks transitions, keys and flow.
- Claude Sonnet 5
- Forced tool use
- Write track notesAI step
Why each track was chosen and how to mix into the next one, written in parallel batches.
- Claude Haiku 4.5
- Review and exportHuman checkpoint
You tweak the order, then export straight back to your DJ software.
- Serato .crate
- Rekordbox XML
- M3U
- App
- Next.js 16, React 19, Tailwind, Vercel
- Data and auth
- Supabase Postgres with row-level security
- AI
- Claude Sonnet 5, Claude Haiku 4.5, Anthropic SDK
- Music data
- Last.fm, Beatport, ACRCloud track ID
- Business
- Stripe billing, Loops email, Sentry
The Tech Stack
I built it full stack from the start, without no-code tools. The full architecture:
Frontend: Next.js 16, React 19, Tailwind CSS v4, shadcn/ui and Base UI, Lucide React
Platform: Vercel with Fluid Compute, Edge Middleware and Node.js 24
AI engine: Claude Sonnet 5 and Claude Haiku 4.5 (upgraded from Sonnet 4.6 in September 2026) with web search, structured tool use and an agent pipeline built on @anthropic-ai/sdk
Data: Supabase with 12 tables, full row-level security, Supabase Auth and @supabase/ssr
Integrations: Stripe, ACRCloud for Track ID, Last.fm, Resend, Beatport and DJcity
DJ-specific: Serato database parser, Rekordbox XML, the Camelot key system, the MediaRecorder API for in-browser recording, @react-pdf/renderer
Going full stack was deliberate. The DJ-specific layer (Serato parsing, Rekordbox export, Camelot key logic and in-browser audio recording) can’t be wired up in n8n. It needed real code, and it’s what makes SetLab useful to a working DJ.
What I Learned
AI handles the mechanical parts of a set and leaves the musical ones to me. When to drop the energy, how to read a room and which record fits a moment are still my calls. SetLab takes care of BPM sequencing, key compatibility and the energy arc, so I can focus on those.
Smaller stages beat one big prompt. The first version asked Claude to do everything in a single call. Splitting it into a blueprint, a selection pass and a notes pass, with Haiku writing the notes in parallel, made sets faster and kept them inside the server’s time limit.
Going full stack was the right call, even though it was slower. Early on I was tempted to prototype with no-code tools, but the features DJs care about most, like Serato and Rekordbox export, Camelot keys and in-browser Track ID, needed a proper build. The extra time paid off in the output.
What’s Next
I use SetLab for my own gigs and keep developing it. Stripe billing is live, so it’s ready for paying users. Next on the roadmap: deeper library management, recommendations that account for the venue, and tighter integration with the platforms DJs already use.
Building something I use myself keeps me honest. After running plenty of my own sets through it, I know what works and what still needs fixing, and that list decides what I build next.
Watch the Full Breakdown
The video above covers the whole build: the problem, the architecture and a live demo.
Follow along on YouTube for more builds like this.
