Fall 2025 Internship @ RVNG

Quick Overview

I returned to RVNG Intl. in fall 2025 to build on the data infrastructure I had developed during the summer.

My work shifted toward analytics engineering and internal data products. I built an easy-to-use ingestion application for the team, developed a catalog analytics layer using Luminate data, and designed time-series metrics for understanding how tracks moved through streaming ecosystems.

The Next Problem

By the end of the summer, RVNG had a centralized foundation for its sales, revenue, and catalog data.

The next challenge was making that infrastructure easier to operate and more useful for analysis.

The team needed a simple way to run ingestion workflows, a reliable connection between streaming records and the internal catalog, and better methods for distinguishing meaningful growth from ordinary variation.

My Role

My fall work focused on three related systems:

An ingestion web application for managing recurring data workflows

A Luminate analytics layer connecting ISRCs, tracks, albums, and artists

A set of time-series metrics for examining catalog momentum, volatility, and long-term performance

Together, these projects moved the system beyond centralized storage and toward reusable catalog intelligence.

Building the Ingestion Application

During the summer, I had built the underlying pipelines that processed RVNG’s data. In the fall, I designed a web application that made those workflows easier to operate.

The application provided a central interface for processing data from sources such as digital revenue reports, physical sales, Bandcamp, and Shopify.

Instead of requiring someone to run individual scripts or manage files manually, the application provided a clearer control point for recurring ingestion tasks.

This project made me think about a part of data engineering that is easy to overlook: a pipeline is only useful if someone can operate it reliably.

Connecting Streaming Data to the Catalog

I also developed an analytics layer using Luminate streaming data.

A streaming record alone does not provide the full context needed by a record label. The team needed to connect an ISRC to the corresponding track, album, and artist before it could understand catalog-level performance.

I created a relationship structure that followed:

ISRC → Track → Album → Artist

This allowed streaming observations to be analyzed in the context of RVNG’s catalog rather than as isolated rows.

The mapping layer also created a reusable foundation for comparing performance across tracks, releases, and artists.

Measuring Catalog Performance

Once the streaming data was connected to the catalog, I explored ways to describe how a track’s performance changed over time.

I developed and tested metrics involving mean performance, standard deviation, relative strength, volatility, momentum, NPV, and winsorization.

These metrics were not intended as abstract financial or statistical exercises. They were different ways of asking practical questions:

Is a track growing relative to its own history?

Is a recent increase unusual or within its normal range?

Is the growth sustained or temporary?

Are extreme observations distorting the trend?

Which parts of the catalog may deserve closer attention?

Rather than relying only on total stream counts, the analytical layer helped characterize the shape and timing of a track’s performance.

Outcome

The ingestion application made the existing data workflows easier to operate.

The Luminate analytics layer connected streaming activity with RVNG’s internal catalog structure, allowing the team to move between ISRC, track, album, and artist-level analysis.

The time-series metrics created a more systematic way to investigate momentum, volatility, discovery, and catalog longevity.

Together, these projects turned the infrastructure from my summer internship into a set of internal analytical tools.

What I Learned

The summer taught me how to build a reliable data foundation. The fall taught me how to turn that foundation into something people could operate and use to investigate real questions.

I also learned that music performance cannot always be summarized by a total stream count. Two tracks with the same total may have very different histories, growth patterns, and audiences.

Analytics engineering, in this case, meant preserving that context while making it accessible to the people responsible for the catalog.

Tools

Python, SQL, GCP, Luminate Data, ETL, Data Modeling, Time-Series Analysis, Analytics Engineering

Made with 💕 in The Office at RVNG Intl. / Mexican Summer, Brooklyn, NY