USE CASE STUDY
Unlock hidden value in logs data to enrich SLOs functionality
How F33 has used Vertex AI’s NLP powerful capabilities to track anomalies in the stream of logs for Nobl9 company?
Nobl9 is an unchallenged leader in multi-cloud/environment SLOs solutions. We have been working with Nobl9 using their product for monitoring AI/ML models. You can read more about this here. This experience made us think deeper about combining SLOs and AI/ML. After a few brainstorming sessions, we came out with an interesting business case.
Our customer Nobl9 wants to help their customers and employees by reading through logs and alerting them in case some unusual system behavior is about to occur. In order to do so, the company needs an automated solution for extracting knowledge from logs and detecting time ranges when systems do not act in a normal, expected way. The company has access to historical logs. They can acquire information about system health status from experts.
F33 has followed its AI/ML Framework to deliver AI/ML solution for Nobl9 company. As a result we helped them to formulate requirements, prepare datasets (we used logs from their product Nobl9) and upgrade their IT infrastructure to be able to communicate with the model.
It was interesting for us to adapt Google’s Universal Sentence Encoder to work with log datasets. Using Tensorflow TFX framework it was fairly easy to build a model that was able to calculate anomaly score for next log in a stream. See the picture below as a demonstration of the working solution.
Models were implemented using F33’s MLOps GCP Platform and the diagram below presents all core components used for this implementation.
Nobl9 has confirmed that the solution brought a lot of benefits to their current product functionalities. The most important was that it turned out to be possible to create an anomaly detection model based on logs that would improve and enrich existing SLOs functionality. This unlocks an ocean of possibilities but most importantly can lead to easier product implementation and also the ability to create SLIs that are not biased by humans. Implementing F33’s MLOps Platform gave Nobl9 standardized and reproducible workflow to experiment and compare different ML models on their log data.