Skip to main content

Introduction to Telepath

Telepath is a machine learning engine that turns database data into prediction APIs without requiring any data science expertise.

Features​

  • No data pipelines or ETLs to manage. Telepath reads training data directly from your database.
  • Machine learning models are auto-generated in minutes using our AutoML engine. Each model is custom, based on your specific data.
  • Models are automatically deployed as REST API endpoints that can make real-time or batch predictions. Fully-hosted, serverless, and scalable.
  • Explainability is provided out-of-the-box, so you can understand why a model made any given prediction.
  • Define your projects in simple declarative code files that can easily be version-controlled in git and collaborated on with other developers.
  • Model Monitoring (coming soon… Telepath will monitor your predictions and notify you when your model performance seems to be degrading and you should consider re-training it)

Getting Started​

Deploying your first Telepath project doesn't require any DevOps or AI/ML expertise. The development lifecycle looks like this:

  1. Create an account at telepath.io (1 minute)
  2. Install the Telepath SDK via NPM (1 minute)
  3. Write some simple code to declare your Project Resources (< 15 minutes)
  4. Deploy your project (2 minutes)
  5. Train your model (~30 minutes of waiting. No work from you.)
  6. Make predictions via API (200 milliseconds)

The entire process, including model training time, takes less than one hour. Best of all, it doesn't require you to deploy any new infrastructure or data pipelines.

How does it work?​

You can think of the Telepath engine as “data in → API out”.

The “data” (called training data) is a set of examples that your machine learning model will learn from. For example, to create a model that predicts the price of a car, you'll need to start with some examples of cars and their prices. This data should be tabular (rows and columns) and stored in an internet-accessible database.

You use the Telepath SDK to define three types of resources:

  • Source - a connection to your database.
  • Pipeline - a query that defines what data should be read form your Source.
  • Model Spec - what you want your model to predict.

These resources serve as a set of instructions that tell Telepath what training data to use and where to find it.

Telepath's AutoML engine will automatically read the training data from your database and generate a custom model. The engine intelligently and simultaneously attempts a wide variety of model training techniques, in order to identify the most effective algorithms to construct your specific model. It's like having thousands of data scientists all trying out different algorithms at the same time.

Once the model has been created, it's deployed as a serverless API endpoint that you can use to make predictions. It'll even tell you which data attributes had the largest impact on the prediction, so you can understand why the model made any given decision.

The entire Telepath process was designed to be fast, so that you can quickly deploy a “minimum viable” model, and then iterate on it. You can easily generate new versions of the model as your training data grows.