Neural Network for Paint Mixing

Machine learning applied to a paint mixing machine to achieve precise color matching.

The project

The client required a system capable of supporting paint formulation, consisting of 13 different mixtures, based on a specified RGB color. A database containing approximately 20,000 formulated elements was available.

First screen Vernice Power
Convex Hull
Paint buckets
Second Screen Vernice Power

A feed-forward neural network system was developed to accurately predict new paint formulas. The system includes an interactive dashboard that communicates with a backend based on a serverless asynchronous queue management system using AWS SQS and AWS Lambda, providing a smooth and responsive user experience.

Events on the SQS bus trigger the invocation of an asynchronous worker (AWS Lambda) that executes the artificial intelligence script and updates a database with the status and details of each executed task.

Requirements

  • Database of 20,000 elements with data preprocessing and augmentation
  • Execution of asynchronous workers via queues
  • Metric monitoring for debugging and observability

Tech Stack

  • Python, Keras (Feed-forward Neural Network)

  • NodeJS + TypeScript (API and worker manager)

  • AWS SQS + AWS Lambda (queue management and asynchronous workers)

  • DynamoDB (data storage)

  • React + TypeScript (web dashboard)

  • AWS Cloudwatch (system usage and internal metrics monitoring)