InsightFinder AI Aims To Give Data Scientist Powers To IT

AI observability can give IT deep insight into AI but is more challenging than observability in other tech areas, says InsightFinder AI’s founder and CEO.

AI observability will be instrumental for organizations to move from AI hype to AI transformation, says Helen Gu, the founder and CEO of InsightFinder AI.

InsightFinder, which was founded in 2016, offers observability into AI systems by detecting model drift, providing diagnostics and performing root cause analysis.

Gu is a computer scientist with expertise in distributed systems and predictive analysis. She put her knowledge to work as a researcher with Google where she worked on her patented unsupervised behavior learning algorithm. She is also a computer science professor at North Carolina State University.

MES Computing spoke with Gu about InsightFinder and how the solution can give data scientist-level insight to organizations, how it benefits businesses and her predictions on what’s next in the AI industry.

Can you speak more on what InsightFinder does?

We provide AI-driven IT observability and AI observability to our customers. We built this AI engine, and you can use this engine to monitor your IT systems. This can be applied to traditional IT environments as well as to new AI systems.

As more people start to use AI—[for example] machine learning to do fraud detection and using large language models for customer service, we can help them monitor those AI models to ensure the AI models produce correct results, not hallucinating. Or [ensure] the AI model is running right.

Speaking of hallucinations, is that a form of AI model drift? If not, what’s the difference between the two?

AI model drift is a very interesting concept. It’s types of anomalies but happens in AI models.

For example, when you go to shop, you purchase something using your credit card. The credit card company will run a fraud detection model to decide whether your transaction is legit or a fraudulent transaction. This decision will be produced by a model with a risk score.

When model drift happens, [with a] normal, benign transaction, the model will somehow give a high-risk score. [And with] a fraudulent transaction, the model gives a low-risk score.

When these things happen, we want to tell people that happened because this can impact millions of billions of people, right? And we want to detect that quickly.

This is called model drift, when your model gives abnormal or wrong answers. [With] model drift we typically talk about more traditional machine learning models when you have a very clear definition of the models, and labels, and what is correct and what is wrong. A risk score is a very clear definition.

Hallucination is a little bit more complicated because [we’re] talking about semantic meanings.

How does AI observability differ from observability in other tech areas?

Traditionally, we just worry about our hardware, and then we add software, but now we have models, we have data. You have more dimensions you need to consider.

[AI observability] is particularly challenging. Most IT operators are trained computer engineers. They understand hardware, code and software, but they are not trained data scientists.

For example, [with] network bandwidth, I see a loss rate is high. I know there’s something wrong. But if I look at AI models, and if I tell you PSI [Population Stability Index] is high or low, if you don’t know what PSI means, it will be hard for you to interpret is this bad or good?

The additional challenge is that to maintain and manage those AI systems, you need to have data science knowledge.

That’s where InsightFinder wants to fill this gap. We ... encapsulate data science knowledge in our product so we can explain to you, OK, you have model drift.

Who is your typical customer?

Our customers are pretty diversified. There are customers who are large Fortune 500 companies. We deploy our AI engine in their data centers, in their AI platforms ... we help them manage all these mission-critical environments.

We have some customers who are small companies. They don’t have a lot of resources to run their companies. Some of [them] are AI companies. We work with an online education company, we help them make sure their service is running and they rely on InsightFinder solely to run their website.

Because we are doing this unsupervised machine learning, we can adapt to different environments.

Can you share some of your predictions or thoughts on the AI space in 2025?

There was a lot of hype in 2024. I think this year people will become more cautious on what business value [AI] can bring. How do you quantify business value?

It’s important to understand how to use AI. AI technology is very powerful. It’s not just simply, throw the data into the model and then you will get the magic correct answer. You need to understand how AI models work in different use cases.

The other thing is also to make sure your model is ethical, correct and responsible.