# Lineup → Spotify Playlist

Bought festival tickets before the lineup dropped. Knew almost no one on the bill.
Instead of Googling each artist manually, I built a tool to do it for me.

Take a photo of the festival poster → OCR extracts the band names → 
Spotify API fetches their top tracks → playlist ready. No manual input.

Built for Lollapalooza 2024. Works for any festival poster.

## The technical decision that mattered

First attempt used a general image vision model — over 8GB of VRAM, didn't fit 
the notebook, too expensive to run in production.

Switched to a text-specialized OCR model instead: higher accuracy on typographic 
layouts at a fraction of the compute cost. Added a correction system for band names 
with more than 3 words — reduced recognition errors by 80%.

The constraint forced a better solution.

## How it works

1. **OCR** — text-specialized model extracts band names from the poster
2. **Correction layer** — fixes multi-word band name fragmentation  
3. **Spotify API** — finds each artist, grabs their top tracks
4. **Playlist** — created and populated automatically in your Spotify account

## Stack

Python · OCR · Spotify API · Docker · Angular · FastAPI

## Architecture

Python microservices with Circuit Breaker and Retry patterns.
Background processing with semaphores — 40% throughput improvement.
Angular Standalone Components for the frontend.