Delivering Wow: How the Miko AI works
When it comes to data science at Miko, we have a lot more ideas than we can practically implement. In particular, with machine learning there are many promising business applications but it's prohibitively time consuming to test them all, especially as a small team. Generally, before using a machine learning model you need to apply time-consuming manipulations to the data so that you can use it to train the model. If we could use the data we have in its raw form, we would be able to test many of these business applications for machine learning much more quickly.
Thankfully, there has been a massive leap in large language models in the last few months, most notably OpenAI's ChatGPT.
We’re always working to build an amazing user experience and we would love to be able to preempt our users’ questions, and proactively answer them, even before they begin to write a message. Imagine if we could use the time series that documents users’ actions within our app, to do this. It sounds like science fiction, but in fact there is a lot of implicit information in the way customers navigate through the app before they ask any support questions.
Even more extraordinary could be a button that collects data from multiple sources (Miko, IOS HealthKit, Fitbit, MyFitnessPal, Whoop, HelloFresh and others) with your permission (of course) and designs what our AI thinks is an optimal diet, sleep pattern and exercise routine for you.
Using traditional methods, this type of information is difficult to make predictions from because it would require weeks of data analysis and manipulation. Even after weeks of work, we still might not be able to find interesting patterns.
It is important to stop now and recognise that none of this can replace medical professionals but that doesn't mean all of its insights are useless, at the very least they are cool.
The Current Miko AI
As mentioned in a previous blog post on APIs and their importance, they are what help power our current AI in association with OpenAI's ChatGPT. We have connected with their API and are using ChatGPT as an AI powered healthcare assistant.
Some of the results have been very interesting. Check out the video below: