Keplr turns datasets into prediction engines.
Start with a labelled dataset
Keplr automatically finds a deep learning model that performs well on your data.
Use your model through a dashboard or API
Try the beta
How it works
Keplr spins up a cluster to explore different model parameters and architectures in parallel until it finds a model that performs well on your dataset.
Keplr then serves your model on dedicated, performant and scalable enviroments. Train, test & use your model with Keplr's API & dashboard.
1. What does "finding a good model" mean?
Keplr produces neural networks that operate on your dataset. Keplr automatically and systematically explores different model architectures, tunes parameters to find a model that performs well on your dataset.
2. How do you know that you've found a good model?
Keplr trains on a portion of your data, and uses the remaining part of your dataset as the testing set. Models are ranked by performance on the testing set.
3. How much does it cost?
When you sign up, you get to perform one model search and access that prediction API for free, hosted on our shared servers. For more than one model, and dedicated servers, see our pricing options.
4. How much data do I need?
The exact size of dataset needed to train a neural network varies from problem to problem. If you have a couple thousand data points, you should give Keplr a shot. If Keplr doesn't spit out a good model for you, you get your money back!
5. What if I don't have enough data?
Keplr is working on features that allow you to aggregate data from public datasets, so that you can design a large dataset that describes your prediction problem. Hold tight!
Testing Mode
1 model
Up to 3000 API calls per month
Developer Plan
$30 per model, per month
Up to 15 models
Dedicated hosting
Unlimited API calls
Enterprise Plan
Get in touch for details
Unlimited Models
Scale clusters for each model as needed.
Hands on support & onboarding.