AI agents are all the rage, a trend driven by the generative AI and large-scale language model (LLM) boom of the past few years. Although it is difficult to get people to agree on what exactly an AI agent is, most people agree that AI agents can be assigned tasks and make decisions, with varying degrees of autonomy. It claims to be a software program.
In other words, AI agents go beyond what mere chatbots can do and help people get things done.
Although it's still early days, companies like Salesforce and Google are already investing heavily in AI agents. Amazon CEO Andy Jassy recently hinted at a more “agent-like” Alexa in the future, one that emphasizes actions and not just words.
In parallel, startups are also raising money from the hype. The latest is German company Juna.ai, which wants to improve factory efficiency by automating complex industrial processes to “maximize production throughput, increase energy efficiency, and reduce overall emissions.” That's what I think.
And to make it a success, the Berlin-based startup today raised a $7.5 million seed round from Silicon Valley venture capital firm Kleiner Perkins, Sweden-based Norsken VC, and Kleiner Perkins chairman John Doerr. announced that it had raised US$.
Self-study is the way
Founded in 2023, Juna.ai is the work of Matthias Auf der Mauer (pictured above, left) and Christian Hardenberg (pictured above, right). Der Mauer previously founded predictive machine maintenance startup AiSight, which he sold to Swiss smart sensor company Sensirion in 2021, while Hardenberg is the former chief technology officer of European food delivery giant Delivery Hero. It was.
At the heart of Juna.ai is helping manufacturing facilities transform into smarter, self-learning systems that can deliver better profits and ultimately reduce their carbon footprint. The company focuses on so-called “heavy industry” – industries such as steel, cement, paper, chemicals, wood and textiles, which involve large-scale production processes that consume large amounts of raw materials.
“We work with industries that are very process-driven, and most of them involve use cases that consume a lot of energy,” Der Mauer told TechCrunch. “For example, a chemical reactor that uses a lot of heat to produce something.”
Juna.ai's software integrates with manufacturers' production tools, such as industrial software from Aveva and SAP, and examines all historical data collected from machine sensors. This may include all measurements of temperature, pressure, velocity, and specific outputs such as quality, thickness, and color.
Juna.ai uses this information to help companies train in-house agents to find the best settings for machines, and provide operators with real-time data and guidance to minimize waste. to ensure everything is running at peak efficiency.
For example, a chemical plant that produces a special type of carbon may use a reactor to mix different oils and perform an energy-intensive combustion process. To maximize production and minimize residual waste, conditions such as the gas and oil levels used and the temperatures applied to the process must be optimized. Juna.ai's agent uses historical data to establish ideal settings and takes real-time conditions into account to help operators understand what changes need to be made to achieve optimal output. It is thought that it will convey.
If Juna.ai can help companies fine-tune their production equipment, they can increase throughput while reducing energy consumption. This is a win-win for both your customer's bottom line and your carbon footprint.
Juna.ai dashboard example. Image credit: Juna.ai
Juna.ai says it built its own custom AI models using open source tools such as TensorFlow and PyTorch. Juna.ai also uses reinforcement learning to train its models. Reinforcement learning is a subset of machine learning (ML) that learns models through interaction with the environment. Try different actions, observe what happens, and improve.
“The interesting thing about reinforcement learning is that it can perform actions,” Hardenberg told TechCrunch. “Typical models only make predictions or sometimes produce something. But they have no control.”
Much of what Juna.ai currently does is more like a “co-pilot,” providing a screen that tells the operator what adjustments to make to the controls. However, many industrial processes are incredibly repetitive, so it helps to allow the system to perform real actions. For example, the cooling system may require continuous minor adjustments to ensure the machine maintains the proper temperature.
Since factories are already accustomed to automating system control using PID and MPC controllers, this is also possible with Juna.ai. Still, it's easy for fledgling AI startups to sell their co-pilots. For now, it's just a small step.
“Right now it is technically possible to run autonomously. You just have to implement the connectivity. But in the end it's all about building trust with the customer,” Der Mauer said. Ta.
Juna.ai co-pilot. Image credit: Juna.ai
Hardenberg added that the advantage of the startup's platform is not in saving labor, noting that factories are already “quite efficient” in terms of automating manual processes. The key is to optimize these processes to reduce costly waste.
“You don't get much out of removing one person compared to a $20 million energy process,” he says. “So the real benefit is, can we increase energy from $20 million to $18 million or $17 million?”
Pre-trained agents
For now, Juna.ai's big promise is an AI agent that uses historical data to tailor itself to each customer. But in the future, the company plans to offer off-the-shelf “pre-trained” agents that require little training on new customer data.
“Building a simulation over and over again can give you a simulation template that you can reuse,” says Der Mauer.
So, for example, if two companies use the same type of chemical reactor, they may be able to lift and shift AI agents between customers. There is roughly one model per machine.
But we can't ignore the fact that companies are hesitant to jump headfirst into the burgeoning AI revolution due to data privacy concerns. Those concerns have been addressed at Juna.ai, but Hardenberg said it hasn't been a major issue so far. That's partly because of control over where data resides, and partly because of the promises it makes to customers in unlocking the potential value of big banks. of data.
“I thought it was a potential problem, but so far it hasn't been that big of a problem because we have all our data in Germany for our German customers.” said Hardenberg. “They have set up their own servers and we guarantee the best security. From their side, they have all this data lying around, but they are not as effective at creating value from it. It was mainly used for alerts, or manual analysis. But our view is that they can do more with this data, which is to build intelligent factories. We can become the brains of the factory based on the data we have.”
Juna.ai has been around for more than a year and already has several customers, but Der Mauer said he is not yet at liberty to reveal specific names. However, all of these companies are based in Germany with subsidiaries elsewhere or are subsidiaries of companies based elsewhere.
“We plan to grow with our customers. This is a very good way to grow with our customers,” Hardenberg added.
With $7.5 million in new money in the bank, Juna.ai has enough capital to expand beyond its current six employees and plans to double its technical expertise.
“At the end of the day, we're a software company, and that fundamentally means people,” Hardenberg said.