• Cerebral Valley
  • Posts
  • IUDEX is AI observability that puts developers first 🎛

IUDEX is AI observability that puts developers first 🎛

Plus: Founder Arno Gau on their state-of-the-art anomaly detection model...

August 20, 2024

CV Deep Dive

Today, we’re talking with Arno Gau, Founder and CEO of IUDEX.

IUDEX is a modern observability platform built specifically for developers and AI-driven startups. Born out of Arno's experience at Scale AI and his desire to improve the often cumbersome process of monitoring and observability, IUDEX aims to make it easier for developers to track, debug, and optimize their applications without the need for complex infrastructure setups. The platform is designed to be non-intrusive, offering streamlined solutions for logging, monitoring, and error detection, tailored for the unique challenges faced by teams working with AI and LLM applications.

Today, IUDEX is gaining traction among developers who need a reliable, cost-effective solution to monitor their applications. With its focus on reducing costs by leveraging advanced OLAP databases and integrating AI for log analysis, IUDEX is positioning itself as a go-to tool for startups and enterprises alike. The platform’s ability to seamlessly integrate with existing AI tools and its open telemetry support makes it an attractive option for teams looking to maintain high-quality observability practices as they scale.

IUDEX has built out a state of the art log anomaly detection model to help developers automatically monitor issues.

In this conversation, Arno shares the journey behind founding IUDEX, the technical challenges they’ve overcome, and what’s next on the roadmap for the company.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Arno 💬

Arno - welcome to Cerebral Valley! First off, give us a bit about your background and what led you to found IUDEX? 

Hey there! My name is Arno, and I’m the Founder of IUDEX. Before founding my current company, I was a staff engineer at Scale AI. While I was there, I worked on two of their SaaS products and helped build out a lot of architecture for our internal tools, as well as our external-facing products. We often ran into issues with monitoring and observability at Scale because the platform we used was really expensive which made the experience frustrating for a lot of developers, including myself. That’s one of the reasons I wanted to dive deeper into improving this aspect of development, which led me to start working on IUDEX about nine months ago.

Before that, I worked at a company called Chronicle, which was part of Google X. We focused on security log analysis, so I was able to bring some of the learnings from building analytical log tools into what we’re doing now. We even got the former CTO of that company on board as one of our advisors.

Give us a top level overview of IUDEX - how would you describe the startup to those who are maybe less familiar with you? 

IUDEX is primarily an observability platform. Essentially, when you're building an application you want to monitor what's happening in production—whether it's tracking user activity, errors, or the number of requests—you need a monitoring tool. IUDEX serves as that monitoring tool, specifically tailored for startup stacks. This is where we see a lot of cool and innovative projects being developed. We focus on tracking things like serverless invocations, calls to OpenAI, and to surface these errors in Slack.

Our goal is to make it easy and intuitive for anyone to open the tool and quickly debug what's happening in production.

Who are your users today? Who’s finding the most value in what you’re building with IUDEX? 

Our target audience is developers, particularly those who want to take more ownership of their observability stack. In larger companies, this responsibility is often split between developers, DevOps aka infrastructure teams, sometimes even a DevEx team. This can make it a complex and time-consuming process to set up everything correctly before you can actually start monitoring your service. Our tool simplifies that process significantly. As a developer, you can just install IUDEX, and you'll immediately see what's happening in your application without needing to deal with any infrastructure configurations.

LLM observability has really taken off as a space in the last 18 months. What sets IUDEX apart from some of the other players in the market? 

I think it's more of a philosophy thing, but in my opinion, LLM observability is kind of a niche area that should fit into a larger ecosystem. That’s how I felt back at Scale. That’s why we focus on general observability while also integrating with other great AI tools. For example, we’ve got a tight integration with Guardrails, a company we are close with. This integration lets you see both the inputs going into your LLMs and the outputs, as well as your server status. You really need both to ensure your app runs smoothly and doesn’t cause issues for customers.

There’s been an explosion of interest in agentic workflows and multi-modal AI. How has that shaped the way you’re thinking about building IUDEX? 

What I find really exciting about this space right now is the combination of two key technologies. First, OLAP databases, specifically columnar databases, have become incredibly fast and efficient at searching. That's what we use, and because of that, we can reduce costs by about 90% compared to traditional observability platforms, which is pretty impressive. And we’re able to do this without sacrificing the quality of the searches you get. This means you can ingest more data without worrying that the logs will disappear soon or that  you're logging too much. It's still good to do that, but it’s not as critical as it used to be.

The second part is using AI or LLMs to pre-analyze your logs. We've developed our own custom embeddings model that we've fine-tuned on top of an existing LLM. This allows us to capture errors and logs, group them together, and make it much easier to surface the important stuff for anyone coming in to figure out what's going wrong. Whether you're looking for anomalies in a new deployment or trying to identify errors impacting a lot of customers, our system makes it easy by doing a lot of the log analysis on the backend.

How do you measure the impact that IUDEX is having on your key customers? Any customer success stories that you’d like to share?  

Ideally, we aim to be as non-intrusive and seamless as possible. A good observability platform should get out of your way rather than create obstacles. We've managed to reduce the onboarding time for new engineers by about 2x. With our platform, you don't have to spend time learning how to use the tool; it's streamlined for standard use cases, so you can jump right in and know where to click and what to look for. This saves a lot of time.

Another interesting use case that we use ourselves and that we’ve seen from a couple of our customers is tracking their customer data directly within our platform. They can follow along with a customer’s journey—seeing what the customer clicks on, what they enter in the search box, and how they interact with the product. This is especially useful with chatbots, where you can see the entire conversation history and identify where things might be slowing down or where odd interactions are happening. I think that's pretty cool.

What has been the hardest technical challenge you’ve faced so far while building IUDEX? 

One of the hardest technical challenges we've faced so far, but also one of the coolest things we've accomplished, is with AI log analysis. We built a state-of-the-art log grouping algorithm, which took a decent amount of work and is something we're really proud of.

However, something that's still a bit tricky is navigating OpenTelemetry. In the observability market, there's a new standard called OpenTelemetry that's becoming increasingly popular. It provides a nice, standard data type for building communication between any OpenTelemetry consumer and event emitter. Since the framework is still relatively new, there are a lot of kinks we’ve had to work out. We've forked their client and made a lot of improvements, but every now and then, odd issues pop up that we have to patch or address. 

It's been a bit challenging, but I encourage anyone interested in observability to check out that project. It's a really cool project—super interesting stuff.

How do you plan on IUDEX progressing over the next 6-12 months? Anything specific on your roadmap that new or existing customers should be excited for? 

One thing we're currently beta testing, which we're really excited about, is using static analysis on code to automatically inject logs and traces without the engineer needing to do it manually. The idea is that it reviews your diff, identifies hotspots in the code, and then suggests where you might want to log or trace. For example, it might say, "Hey, you should log this function," or "This function has a lot of high branching factors—you should probably look at the arguments going into it." I think that's pretty cool because it can help solo developers or smaller teams maintain strong logging practices as they scale, without having to constantly think about it. It's like having a senior SRE reviewing your PRs, even if you don't actually have one on your team.

We're also pretty excited about using a similar technique to do a first pass bug fix. We can follow the code path and identify where a bug in the code might be triggering an error and suggest a change to fix it. It only works with simple stuff right now but we're excited because it's been getting better really fast.

Lastly, tell us a little bit about the team and culture at IUDEX. How big is the company now, and what do you look for in prospective team members that are joining?

We're currently a team of five people, but we're looking to grow, so if you know any infrastructure engineers who are on the job hunt, definitely send them our way.

In terms of team culture, we're committed to maintaining a really high engineering bar. We're focused on finding people who not only understand good engineering practices but also genuinely care about implementing them. Since this is a pretty technical product, code quality is really important to us. There's a lot we need to build, so we're always excited to find engineers who can think outside the box or take unconventional approaches to help us reach our goals.

For hiring, we're looking for people with solid infrastructure backgrounds who care about code quality and thrive in a fast-paced startup environment. What’s exciting about working here is the variety of code we get to work with—different languages and frameworks—because we have to support so many. We’re particularly interested in people who are comfortable picking up new technologies and integrating them quickly.

Anything else you’d like people to know about what you’re doing at IUDEX? 

We're always looking for developers interested in trying out our product. Our goal is to make it as easy as possible, so we really value any feedback. Right now, we're offering a special free tier—100 million log lines per month for free—for anyone who reads this newsletter. We know observability often becomes an afterthought, only gaining attention when a serious issue or incident arises. That's when you wish you had a place where all your logs were stored and easily accessible. We encourage developers to think about observability upfront—or at least let us handle it for you. We're here to help you build enterprise-grade software.

Conclusion

To stay up to date on the latest with IUDEX, learn more about them here.  

Read our past few Deep Dives below:

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.