TechPrufer
Get started
Spotify
Spotify logoSpotify · Feature trail

Wrapped Animation System

4 posts · Jan 2020 – Feb 2026

What this trail covers

Spotify's Wrapped experience continues to evolve, with the 2025 iteration introducing the 'Wrapped Archive' feature. This system identifies up to five 'remarkable days' from a user's listening history, generating personalized, LLM-generated narratives grounded in actual listening data. The system uses a priority-ordered set of heuristics to identify these days, such as 'Biggest Music Listening Day', 'Biggest Discovery Day', and 'Most Unusual Listening Day'. A distributed data pipeline aggregates candidate days, and reports are pre-generated asynchronously. Prompt engineering, model distillation, and a fine-tuned production model are used to generate over 1.4 billion reports. The system is designed for high-throughput writes using a column-oriented key-value database and proactive scaling for the global launch. An automated evaluation framework with LLM judges assesses report accuracy, safety, tone, and formatting, feeding into a remediation loop to fix issues and update guardrails.

All Spotify trails
  1. Inside the Archive: The Tech Behind Your 2025 Wrapped HighlightsThis post details the engineering behind Spotify's 2025 Wrapped Archive feature. Key contributions include: the design of a heuristic-based system for identifying 'remarkable listening days' from user data; the use of a distributed data pipeline for aggregating candidate days and object storage for storing listening history; the development of a two-layer prompt engineering strategy (system and user prompts) for LLM-generated narratives; a model distillation pipeline to transfer quality from high-performance models to a smaller, faster production model, enhanced by Direct Preference Optimization (DPO); the architecture for generating ~1.4 billion reports asynchronously via a pub/sub message queue; a column-oriented key-value database schema designed for safe, concurrent writes of user reports; proactive scaling and synthetic load testing to handle the 'big bang' global launch; and an automated evaluation framework using LLM judges to assess report quality (accuracy, safety, tone, formatting) and a remediation loop to address failures.
  2. Exploring the Animation Landscape of 2023 WrappedThis post details the evolution of Spotify's animation strategy for the Wrapped campaign, introducing Lottie as a primary tool for generic and complex animations to achieve cross-platform parity and reduce engineering effort. It also highlights the continued use of native animations for data-driven and interactive elements, outlining the decision-making process for choosing between Lottie and native builds based on parameterization needs and user interaction.
  3. The Audio Aura Story: Mystical to MathematicalThis post details the technical approach to generating the 2021 Wrapped Audio Aura. It outlines a two-step process: first, identifying the top two mood categories and their weights from user listening history by aggregating track moods into six broad categories and then narrowing them down. Second, translating these moods and weights into visual colors using a predefined color mapping and contrast levels based on mood weight. The visual itself was engineered using a combination of animated gradients, ribbons, and blur effects, leveraging existing assets from previous Wrapped campaigns to meet tight deadlines.
  4. Spotify Unwrapped 2019: How We Built an In-App Experience Just for YouThis post details the engineering effort behind transitioning Spotify Wrapped from a web-only experience to a native in-app experience for iOS and Android in 2019. It highlights the development of a real-time rendering backend service using CEF, HTML templates, and CSS, the challenges of global localization and handling diverse user data, and the system architecture changes including GKE scaling, caching layers, and image generation for load handling. The post emphasizes a lean, MVP-driven development approach and the resilience required to manage last-minute issues.
  5. Spotify Unwrapped: How we brought you a decade of dataThis post details the engineering challenges and architectural decisions made for the 2019 Spotify Wrapped campaign, focusing on processing a decade's worth of user listening data for over 248 million monthly active users. The User Wrapped data engineering team processed approximately 5x the amount of data compared to the previous year at a fraction of the cost. Key architectural decisions included using Google Cloud Bigtable for efficient time-series data aggregation and as the final data store to reduce shuffles. The system decoupled data processes into smaller 'data stories' for faster iteration and implemented a Python library for direct data access from Bigtable to ensure data quality. The primary takeaway is the importance of leveraging the right system design and data store to reduce cost and decoupling data processes to improve iteration speed.