Feb 6, 2024
Alberto Espinosa
How do we make recommendations?
In MilkStraw AI, we base our recommendations on agents, these agents are neural networks that are trained on historical cloud data. They are trained on those datasets and as they adjust the parameters, we expect them to give us the best possible optimization recommendations as an output.
Data is an important part of this process. Datasets that contain more quality & quantity deliver better recommendations, but more importantly, is it possible to access high-quality data without invading your privacy?
The answer is YES. Let me explain how this happens behind the scenes.
Behind the scenes
In the first stage, we start by extracting data from the EC2 APl. We want to know your usage, resources, types, and families of instances you use, and how long your instances were turned on.
Then we feed it into our general model which was trained on open data sources. But to prepare our agents to handle black swan events, or in other words, “what if” situations, we enriched them with synthetic data.
Synthetic data is a very important part of how we train our models. We use it on open data sources and your data because we want to ensure high encryption of your data by anonymization and we increase the accuracy of our recommendations by generating the “what if” scenarios mentioned above.
After we feed your data to our model and fine-tune it, we start populating the recommendations on your dashboard. You can rest assured that those recommendations come from vast scenarios our models are trained on and your own utilization data, puts the cherry on top.
Check your recommendations today