Begin AI Personalization Tech

 
Personalization is at the heart of every good product experience, and by personalization I refer to improving product experiences to tailor them exactly to what the user might be interested in. Some examples of personalization are:
  • Netflix recommended movies
  • Wearables offering personalized coaching
  • An app that generates videos in a person’s voice (deep fake)
  • A personalized diet app based on health profile, eating and exercise habits
  • A future personalized vaccine based on a person’s genetics
 
What is in common between all these use cases is that they require learning from a large sample of people, their interactions and their responses to recommendations (what we may call trials) to finally develop the wisdom that allows a company to predict with a certain accuracy the right positive personalized experience.
 
Running these trials and sampling traditionally requires a few things:
  • Tracking user interactions and behaviour
  • Copying that tracking data to central servers where machine learning algorithms live
  • Running heavy duty machine learning models on all that data
  • Trying out the results on new users
  • Rinse and repeat until desired outcomes are reached
 
The above lifecycle is complex, expensive and very error prone. It’s like having to fish the whole ocean to find out if your fishing technique is correct, and then try again.
Key challenges in this process are:
  • Bias. You have no known method of knowing how skewed is your data until you’ve put it live, a few months later which can really harm a business. Imagine recommending red tshirts only to all people. (exaggeration for the purpose of explaining the point).
  • It takes science, engineering and business to work very hard for a very long time together only to setup. And then business might find out that this was the wrong experiment, but because they invested so much in it, they have to defend it.
  • Privacy. Because you don’t know what data you need, you start piling up any data you can find; which if hacked can have catastrophic effects - as we keep seeing in the news.
  • Not enough data. Even if you piled it all up, you still may not have sufficient userbase to develop that deep understanding of your users.
 
These challenges motivated us at Begin to reimagine a general purpose personalization infrastructure that solves for the above challenges and allows for more businesses and scientists to build awesome new personalization experiences that truly augment the personal needs and desires of their audiences.
 
We created an end to end personalization platform that allows for machine learning algorithms to learn from data in-place without moving it to the cloud making it fast and easy to develop crowd wisdom with little to no technical setup.
 
Here is how we do it:
We only move to a cloud operated by Begin a tiny mathematical signature of the data that is sufficient for Begin’s cloud artificial intelligence engines to extract meaningful user insights and enhance any given personalization algorithms. With this:
  • Begin’s signatures are tiny and fast to generate, so you can run as many experiments as you like within minutes instead of months.
  • Begin’s signatures are packed with information, yet are anonymous and reveal no personal or private information about the user making your overall system secure yet highly performant.
  • Begin’s cloud AI learns very fast, and dynamically adapts these signatures to your objectives, making it possible to learn for a diverse set of use cases without modifying any of your data infrastructure.
 
With that, Begin is an end to end ML infrastructure solution that allows you to learn fast and adapt from your users without setting up any pipelines or any data storage. It’s completely serverless and requires no data setups from your end.
To use Begin all you have to do is perform three steps:
  1. Setup the SDKs on the devices that generate the data, currently we support Android, iOS and Python SDKs.
  1. Write a json data schema and provide it to Begin interface, which will be used to orchestrate the algorithms on the devices.
  1. Start experimenting, using either ready made algorithms provided by Begin team or building your own algorithms. In either case our deep personalization engines will optimize the outcomes for you.
 
FAQs:
  • Do all devices need to be online at the same time for the learning to happen? No. the systems are 100% asynchronous, and learning is continuous. Whenever devices come online, their learnings join the crowd.
  • Is this technology an old trick? No. an invention we worked on for a couple of years and it is patent pending.
  • Is this federated learning? Inspired by it, we started building federated learning systems then we found a better way to do this and our techniques are novel, designed to be simple, safe and fast.
  • Can this technology be used for multiple apps to cross-learn from each other? Yes. that’s a feature, you can have a diet app cross learn from a cycling app using our orchestration platform without any custom work that’s because our tech is built as decentralized anonymous learning platform from the ground app, so you can share learning without sharing data. If you’re interested in this use case, please give us a shout!
  • What if my data is poor or I don’t have enough users? We are developing some zero-shot transfer learning based models that you can use until you have data.
  • This sounds so simple, why hasn’t it be done before? Building simple solutions takes a lot of hard work 😊