PacketAI predicts IT incidents by parsing large event data sets

Romain Dillet

Meet PacketAI, a French startup that wants to alert you when there’s something wrong with your app or service. The company uses machine learning to parse raw event data and find out if there’s anything wrong.

PacketAI can intercept incidents at many different levels. For instance, the service can tell you if your users can’t write something on your database or if there’s something wrong with your compute layer.

PacketAI doesn’t try to reinvent the wheel. The startup is well aware that there are many monitoring tools out there — Datadog, Splunk and Dynatrace for instance.

“Those tools are primarily designed for humans so that they can understand information delivered by machines,” co-founder and CEO Hardik Thakkar told me.

PacketAI integrates directly with the APIs of Datadog, Splunk or Dynatrace to analyze raw event data in real time. Instead of scrolling through thousands of lines, you can get an alert that tells you that bank transfers take a lot more time than usual to go through, for instance.

Eventually, you should be able to repair your problem much more quickly, which could potentially improve your revenue.

For now, the startup creates a machine learning model for each client. But the plan is to create a model for each vertical as soon as you have four or five companies in the same space using PacketAI. You could imagine a model for banking companies, a model for telecom companies, etc.

The startup already raised $2.3 million (€2.1 million) from Aster Capital, BNP Paribas Developpement, Entrepreneur First and SGPA.

PacketAI is already working with some clients on the first implementations of its product. The service will be available to anyone in early 2020. Pricing varies depending on the number of nodes (any physical or virtual network element) you want to monitor using PacketAI.