A machine learning platform using pre-trained genetic algorithms. Data-set agnostic.

It's not rocket science. It's data science.

Our software is available in a variety of flavours, a SaaS platform hosted on IL0 to IL6 infrastructure in the geographical region of your choice, through to an OEM IP based solution which can be applied directly to your embedded solution.


SaaS Platform (accessed by a RESTful JSON API)

Our Machine Learning SaaS Platform learning algorithm has already been trained on a variety of problems and cohorts of data. To generate an inference on the unseen data a JSON object is sent to the relevant endpoint as so.

https://client.yellowsubmarine.io/api/inference/prod

State of the art Model Builder, using unique algorithms for Feature Selection and Genetic Linear Regression, utilising massive Google Firebase backend non relational JSON database to store blobs.
With a maximum of 1,000,000 connections and almost unlimited data persistence, you will be able to build your Big Data predictive models.



Our model builder uses Googles Firebase non relational database at its back end, allowing over 1,000,000 connections and almost unlimited storage for your massive data sets to build models.

After our patented Feature Selection algorithm removes features that are not predictive, model builder will crunch the numbers - a 100MB, 250,000 row and 10 column data set will take around 4 hours to compute.

Small code lightweight footprint for inference engine

Our Inference engine is very lightweight as once the algorithms variables have been calculated by our model builder, around 10 lines of code are needed to complete an inference on unseen data for our genetic linear regression algorithm. We are currently demonstrating our software on desktop and mobile processors, including ports to embedded C. Because we have put all the clever stuff in the model builder which does all the number crunching the standalone inference engine is encoded with the results.


Applications

E-Commerce and Online Shopping  


Using our models you can quickly determine whether a shopper is going to abandon their shopping cart well before it happens, without needing annoying pop ups. As soon as your customer goes onto your website by presenting our API with your customer information we will give you an inference telling you if you should offer them a discount then and there.


IoT and Radio Communications Edge based applications

By building offline models that have been trained on the Tera byte or even Peta bytes of data involved in radio comms such as 4G and 5G, but with a separate lightweight inference engine and therefore small amounts of compute and can be hosted on an existing ARM stack. For applications requiring fast inferences (such as sensor applications like Wind Turbines) our data models can be loaded on the Edge of the device. F

Computer Vision, Face Recognition and Voice Recognition

Our models can be trained on a multitude of datasets, demographics and situations. Again because the inference engine is separate to the model builder, our offline models can be hosted in very compact code on the edge.


Medical Diagnosis and Recommendations
 

Data privacy and patient confidentiality is a major issue when creating predictive models. Once our models are trained on your data set (again you can do this in your own privacy using our model builder), then once you have encoded the algorithm values into the inference engine you don't need any confidential patient data for the system to work.

Video, Audio Compression and Storage

Using our machine learning models we can drastically eliminate wasted storage. In a world where everything is being reduced apart from data storage our software will enable you to reduce your storage footprints from Petabytes to Terabytes. Think about all those emails people send with massive attachments that can be identified the same, or data that is invariably never accessed and could be moved off and archived. Our algorithms can be trained on datasets that are individual to you.

Credit Risk

With a whole new wave of alternative lenders and challenger banks disrupting the banking and lending industries, a new and fairer way to identify if a consumer is a credit risk is required. Basing future outcomes on past historic data is by far the best way in stationary datasets.  By using our unsupervised machine learning models we can increase your pass rates and reduce your defaults. Also using our Feature Selection Tool it gives you insights in easily readable human language with our Magic Donut view.


Intellectual Property


Working with us you can guarantee that embedding or using our software you have all the necessary rights and licenses. We take our intellectual property very seriously. All the ideas for our software and hardware have been originated by our founders, or we have properly licensed the relevant IP and to give us a priority date for our IP various patents in multiple vertical industries were registered over 1 year ago, GB1819646.9, GB1819645.1, GB1819644.4, GB1819643.6, GB1819642.8, GB1819641.0.

Thanks! You have been added to our beta list. 

Help spread the word!

Oops! Something went wrong while submitting the form. :(