Automation of Sorting Collectible Cards

There are thousands of types of Magic: The Gathering cards with different attributes (characters, abilities, colors, etc.). The cards are randomly packaged, as the game is also collectible. The client trades cards — buying them from people who already own them and selling them to those seeking specific ones. For this to work, cards must be manually sorted and cataloged. Each year, this involves millions of cards, requiring a team of about 10 full-time employees. When fulfilling an order, it is necessary to check card availability in stock, locate them, and prepare them for shipping.

We designed a card-sorting machine and developed an algorithm that converts a scanned image of a card into a hash and compares it with hashes in an open-source database. At the same time, it stores the history of individual runs with recognized cards and their metadata on the backend. On the frontend, the client has an overview of runs with basic statistics (number of cards, time, accuracy), a list of recognized cards, and a list of those not recognized with sufficient certainty and awaiting manual verification — the system suggests possible matches in an “autocomplete” style.

Buyers

Google Cloud IoT Core, Redis Enterprise Cloud on Google Cloud (with PostgreSQL), Google Cloud Storage, BigQuery, Google Cloud Functions, Google Compute Engine, Cloud Run Admin API, Docker image creation, Raspberry Pi

Head of business unit

Full implementation

Reduction of repetitive work

Revolt BI oversaw the development of both hardware and software for the automatic recognition and sorting of collectible cards. The image recognition algorithm in Python used machine learning models trained in TensorFlow. The backend architecture was designed and operated on Google Cloud Platform.

The greatest contributor to the project’s success was the close collaboration between the technical and business teams, which enabled quick responses to client needs and iterative fine-tuning of the solution.

We recommend others to begin with a thorough understanding of the domain and user needs — in our case, it helped that the team understood how the collectible card market works. We also recommend building the solution modularly and scalably, leveraging cloud services that allow for easy automation and process adjustments. Close cooperation between technical and business teams is crucial — regular communication prevented misunderstandings and accelerated development.