Anarion I is the name of the product we are building for NextGenInterview (NGI). There are two elements to this product:

1. The machine learning algorithm itself.
2. The task of integrating it within NGI’s application programming interface (API).

Speaking to Nick, it is understood that there are two methods of accomplishing this second step.

1. Allowing the API to remotely access Anarion I’s code, (for instance through our own API) which will be stored in the cloud.
2. Providing NGI with Anarion I’s code directly which they will include in their own API’s code.

Going for 1. would be the better option: Working with API’s will become essential anyway, so picking up those skills now would just kick start the learning curve. 1. also sounds more difficult.

What the f*%$ is a machine learning algorithm?

It is computer code which ‘reads’ the data and outputs numbers. It is written in the programming languages Python or the programming environment R. The models underlying the code developed out of research into how the mind functions. Here is an example:

• NextGenInterview allows recruiters to video interview candidates and then review these videos later. It’s complementary to a CV/text based application. Importantly, CV’s and text answers to questions are provided by candidates also.
• Anarion I will look at the text provided by the candidate (CV and question answers). Their answers will then be compared to exemplary answers provided by the hiring firm. The candidate can then be scored based upon correlation between the two sets of text.

How does Anarion I work?

It is based upon the ‘random forest’ algorithm. This is basically a massive flow chart, composed of a multitude of ‘decision trees’. The exact number of branches, and the placement of branches is actually devised by the algorithm. Skill however is required in ‘pruning’ the forest and in optimising the model.

The way that machine learning works is through a process called cross-validation. What happens is that the random forest learns from practise data and then has the job of trying to predict from test data.

So Anarion I could work by scoring the candidates in terms of how likely it thinks the candidate is to be hired. And we could compare the algorithm’s answers to who was actually hired. Does it predict correctly? The best bit is that the learner can improve over time as the volume of data increases.

Any questions about machine learning / artifical intelligence ? Post below. I’ll answer.