clanker-fm is an ai-powered bot that judges your spotify taste.
you log in, clanker digs through your listening history, asks you a few pointed questions, and then delivers a roast you didn’t ask for but probably deserve.
it also connects you with other users who share your questionable taste.
| Category | Highlights |
|---|---|
| 🤖 AI Roast | Interactive Q&A + instant AI reactions → final teardown, roast score, shareable card |
| 📊 Metrics | Variety score, basicness index, genre diversity, and listening pattern analytics |
| 👥 Neighbours | Similarity search (weighted Jaccard) → ranked matches, “mutual bad taste” callouts |
| 🏠 Dashboard | Terminal-style homepage with top artists/tracks/genres → fast, cached, responsive UI |
| 🧩 Extras | Export system (html2canvas), edge-cached APIs (<100ms), retry states, fully responsive |
| Category | Technology |
|---|---|
| Framework | next.js 14 + app router |
| Auth | nextauth.js (spotify) |
| Styling | tailwind css |
| Animations | framer-motion |
| AI | openai api / gemini api |
| State | react context + hooks |
| Export | html2canvas |
| Database | supabase postgresql |
| Deployment | vercel |
clanker doesn’t just spit out a verdict — he makes you squirm first.
- scanning recents: clanker looks through and roasts your recent played music, while processing the rest of the data in the background.
|
| step 1: scanning your spotify recently played |
-
interactive q&a: 6 multiple-choice questions based on your own data (“which of these tracks screams ‘me, but a more annoying version’?”)
-
live responses: clanker reacts instantly to your picks with dry, surgical commentary
|
| step 2: interactive Q&A with snarky responses |
- final analysis: variety score, basicness index, most replayed track, “tracks of concern,” and emotional support artists
|
| step 3: the final roast verdict and scores |
- verdict: a paragraph-long teardown of your taste, equal parts psychoanalysis and insult comedy
- rating: a brutally honest score out of 10 (margin of error: 0)
- exportable: save the roast card to share your shame
|
| exportable custom roast card |
because numbers hurt more than words.
- variety score — how adventurous you are (or aren’t)
- basicness index — a polite way of saying “you listen to the same stuff as everyone else”
- genre diversity — counts how many genres you actually touch
- listening patterns — time-based habits, peaks, and troughs
clanker will happily point out who else is stuck in your sonic bubble.
- similarity search: weighted jaccard algorithm compares your top artists, tracks, and genres to other users
- ranked list: most compatible “victims” shown with similarity %, common artists/tracks/genres
- mutual bad taste: if you share questionable habits, clanker will call it out (“you both have terrible taste in experimental hip hop and alternative hip hop”)
- shared delusions: lists the genres you both cling to
- db-backed: powered by supabase postgresql so the database is updated in real-time every time someone signs up
|
| find your musical neighbours (and mutual bad taste) |
the calm before the roast.
- terminal-style dashboard with your top artists, tracks, and genres
- responsive and cached for instant reloads
- persistent nav tabs so you can jump between sections without losing your place
- clean, minimal, and just a little smug
|
| listening metrics dashboard |
things clanker does because it can.
- export system via html2canvas (stats + roast + album art)
- edge-cached api routes for <100ms responses
- graceful loading + retry states (clanker doesn’t panic)
- mobile-friendly + fully responsive because most people will probably run this on their phones
- light/dark mode support for inclusivity
 |
 |
|
| mobile dark | mobile light | exportable roast card |
 |
 |
| light mode | dark mode |
spotify wrapped is fine, but it’s once a year and has become way too boring and bland. i wanted something that:
- was unserious and crude
- tore your music taste apart
- made me question my life choices
- worked whenever I felt like it
- could share and connect people with similar tastes
- doubled as a playground for ai, similarity search, and rapid ui iteration
- also, clanker needed a new hobby.
a few years back i remember coming across pudding.cool, which did a viral spotify “roast” —
funny, but it was static and rule-based (no real ai).
clanker.fm cranks it up:
- actual ai roasts (openai / gemini)
- interactive q&a tied to your own data
- live commentary + neighbours with “mutual bad taste”
pudding walked so clanker could roast at scale.
- spotify oauth via nextauth.js
- ai roast generation (openai api and gemini api options) with interactive q&a
- fetch top tracks, artists, genres, patterns
- homepage sections populate with stats
- neighbours query supabase postgresql db + compute similarity (weighted jaccard algorithm)
- export custom roast card via html2canvas
- local caching for roast, stats, neighbour matches
- responsive design from mobile to desktop
- edge-cached api routes for speed
- minimal loading states, no unnecessary spinners
clanker used to run on multiple prompts and API calls — one for scanning comments, another for the Q&A, another for the final verdict, plus a few extras.
this meant:
- repeated injection of the same spotify data
- higher token usage (and cost)
- more latency between roast stages
- more places to update when tweaking tone or style
- for the interactive Q&A, 6 separate API calls — one per question — slowing things down
now it’s down to just two prompts — one for scanning comments, one for the complete roast experience — and it precomputes all answers for the interactive Q&A in a single call, so the rest of the flow feels instant.
| Before | After |
|---|---|
| 4–5 separate prompts & API calls | 2 unified prompts |
| repeated data context in each call | single data pass per stage |
| higher token usage & cost | fewer tokens, cheaper runs |
| Q&A: 6 API calls (one per question) | Q&A: precomputed answers in one call |
| more latency between steps | faster, smoother roast flow |
impact:
- ~40% fewer tokens per roast
- noticeably faster response times
- instant Q&A interactions (no waiting between questions)
- simpler, easier‑to‑maintain prompt definitions
git clone https://github.com/hritsh/clanker-fm.git
cd clanker-fm && npm install
cp .env.example .env.local
# add spotify creds, ai keys, db url
npm run dev- spotify developer app
- openai or gemini api key
- nextauth secret
- supabase postgresql db keys
MIT
|
built by hritish | source on github "your playlist probably sucks" — clanker |
|
+---+ |o_o| |_-_| |