AI* MADE EASY

(*deep learning for business data)

We’re taking the OSS route

Let’s not kid ourselves., building ML/AI models is hard, putting them in production is harder and building a business around them is the ultimate challenge. We’ve spent the past 3 years on solving that challenge and building a deep learning platform as a service (AI/DL PaaS) for customer-related data, which enables businesses to predict their end customers' behavior and act based on those predictions proactively, rather than reactively.

Now we want to make it available as an Open Source Solution.

COMING UP

airt-keras
The extension of the Keras framework with a set of new layers specialized for business data such as transformers, time-embeddings, monotonic layers, etc., and with additional classes for training them such as LRFinder and OneCyclePolicy. These layers can be used to easily build SOTA models as we already demonstrated.

airt-service
Orchestration of training/retraining and prediction of custom models written in Python language, regardless of the ML framework used. Behind the scenes, the service builds Docker images running the models and uses AirFlow to schedule recurring operations or execute them on demand. We already support running ML loads on AWS and Azure, and we will add support for other cloud providers in future.
The REST end-points for each method are being generated together with appropriate user management and restrictions. We plan to integrate such REST endpoints into different marketplaces (AWS, Google cloud, Shopify, Alibaba, etc.) in order to shorten the time spent on monetizing the model and receiving feedback from the real customers.

airt-client
A system for publishing pip and conda packages with python library and CLI commands for accessing a particular API for the end users. With such systems, a developer can quickly generate a website with landing pages, tutorials and technical documentation, together with CFA that would help monetize the API.

COMING SOON

open-embeddings
Integration of open data in models is very hard today. In talks with potential clients we got very positive feedback on the idea of using embeddings of time and place to augment their data in modelling. For example, tabular data from EuroStat can be used to embed cities in time and then use that information instead of sensitive personal information of each client. This project is under way and we plan to have the first system working by the end of 2022.

DRIVING BUSINESS IMPACT WITH DEEP LEARNING

We focus on event-based data that dominates all customer-related business applications. Doesn’t matter what you want to predict - which customer to target or who is about to churn - we got you.

BUILDING A LOW/NO CODE “AI BRIDGE”

We want to empower data scientists, but also enable developers to use AI, allowing them to use deep learning without having to learn it.

BLEEDING EDGE TECHNOLOGY

We invested a lot into developing deep learning techniques that can extract the most out of customer data. Our solution is outperforming - in terms of accuracy - all other alternatives on the open datasets (we have SOTA results against Google and IBM.

airt-client sneak preview

See docs