Makes interface easier to use with customer support chatbot powered by Gemini 1.5 Flash
Helps users generate code using natural language with coding assistant built with Gemini 1.5 Pro
Increases accuracy of simulations with anomaly detection tool
Saves customers time and money on simulations with a range of generative AI solutions
Expands customer base by making simulation services more accessible
Multiphysics simulation platform Quanscient wanted to make it as easy as possible for users to run simulations. With Gemini it’s building a range of generative AI solutions to make its software easier to use, helping users get more accurate results, more quickly.
Multiphysics simulation platform Quanscient wanted to make it as easy as possible for users to run simulations. With Gemini it’s building a range of generative AI solutions to make its software easier to use, helping users get more accurate results, more quickly.
From electric vehicles to fusion energy reactors, the only way to know if a design will work is to test it. Physical prototypes take time and money to build, so instead scientists and engineers often use multiphysics simulations to understand how their designs would behave in real-world conditions. This lets them hone in quickly on viable designs, accelerating their development cycle.
Traditionally, multiphysics simulations use compute-heavy on-premise simulation machines, which are both slow and costly to use. As a result, they’re very limited on the complexity of simulations and the number of simulations that can be run. Enter Quanscient. With its cloud-based simulation platform, the company is helping its customers increase their simulation throughput: engineers can run thousands of complex simulations parallelly in the cloud to efficiently explore different designs. With Quanscient, any company can subscribe to its browser-based platform, define its test using either a graphical user interface or a coding API.
While Quanscient’s platform is easy to access, its graphical user interface previously required a certain level of knowledge to use it effectively, a fact reflected in the number of queries its customer support team received.
Gemini gave us very accurate results. Because of its huge input-context window, we could simply dump the entire code base into Gemini to give it the necessary contextual understanding. That high input token limit meant Gemini was the only LLM we could use to build our coding assistant.
Çağlar Aytekin
Lead AI Developer, Quanscient
The coding API, meanwhile, required a strong understanding of Python to design accurate simulations. With the emergence of generative AI, Quanscient saw the opportunity to make both its user interface and the coding API more user friendly. It began designing a coding assistant to support customers using its coding API, and set about testing a number of gen AI solutions. The company soon found that due to the sheer size of its code base—a document running to hundreds of pages—Gemini from Google Cloud was the only technology it could use to build a workable solution.
Making coding simple, with a coding assistant
Quanscient uses Python API calls using Vertex AI to access Gemini. According to Quanscient Lead AI Developer Çağlar Aytekin, “Gemini is highly intuitive and very straightforward to use,” and it took the Quanscient team just two to three weeks to learn its way around the family of models. Quanscient’s coding assistant is now in the demo stage. Built using Gemini 1.5 Pro, which excels at advanced coding tasks, the solution will soon make it far easier for Quanscient users to design simulations with code.
“Our scripting interface can be hard to use for first-time users,” says Aytekin. “With this coding assistant, users will be able to use natural language to describe the kind of simulations they want to build, and Gemini will generate the lines of code for them.”
Supporting users through the design process with a gen AI chatbot
While Quanscient’s graphical user interface requires less knowledge to use than the coding API, the Quanscient team wanted to make sure that this, too, was as easy to use as possible. Looking at users’ interactions with its customer support team, it noticed they were often asking questions about how to use the interface to design their simulations. The Quanscient team realized that giving users a chatbot to support them as they designed their simulations would help them find answers to their questions as they went along.
Much of the documentation explaining how to use the Quanscient platform involves textual explanations supplemented by diagrams and illustrations. This meant Quanscient needed a generative AI model capable of reading both text and images. As a natively multimodal family of models, Gemini was the perfect fit. Quanscient chose Gemini 1.5 Flash to build its chatbot, which is now deployed, enabling Quanscient users to ask natural language questions about how to design their simulations. The chatbot searches through the text and images in Quanscient’s documentation for the answer, then gives the user a natural language response explaining exactly what to do.
We want to make our platform as fast and smooth as possible. With Gemini we have built a chatbot that lowers the barrier to entry to using it. It also takes the weight off our customer support team’s shoulders, which was one of our main motivations for building it.
Çağlar Aytekin
Lead AI Developer, Quanscient
Reducing the risk of failed simulations with anomaly detection
Having found a way to help customers use its platform more easily, Quanscient also wanted to help users ensure their designs were accurate. A flaw in a simulation design can result in that simulation failing altogether. With Quanscient’s pay-per-use model, that meant that a single flaw—even just one misplaced zero—could lead to a user wasting their money on a failed simulation.
To fix this, Quanscient designed an anomaly detector using Gemini 1.5 Flash. Soon to go live, this solution will use its knowledge of previous simulations to detect when there is a fault in a user’s completed simulation design. “Now, before pressing the ‘run’ button, users will be able to press a ‘verify’ button, which will send the user’s design to Gemini to check it for flaws,” Aytekin explains. Errors will immediately be flagged, allowing the user to correct them, saving them the cost of a wasted simulation, as well as the time spent trying to figure out what went wrong.
Predicting patterns and saving time with a geometry selection tool
Given the complexity of the geometrical designs that users create on the Quanscient platform, there are parts of the design process that are very intricate and take a long time to perform. In particular, the Quanscient team noticed that when users wanted to assign a specific physics rule to a lot of different points on their geometrical design, it was taking them a long time to do, as they needed to perform the process by hand. This also meant the process was prone to error. Now, Quanscient is using Gemini 1.5 Flash to develop a proof of concept to remove the manual part of this process and automate it instead.
“Let’s say a customer is designing a computer chip,” explains Aytekin. “There are so many small electronics components involved, assigning a particular voltage to each one would take a long time by hand. Our solution will use Gemini to recognize the pattern of a user’s first few selections and then automatically select the remaining points on the geometrical design. This will save the customer a lot of time, and make the process far less prone to error.”
Envisaging a simpler future with fully automated text to simulation
As Quanscient continues to develop and roll out these solutions, it’s actively scoping out new ways to use generative AI to support its customers, make its services more accessible, and increase its customer base. Ultimately, Quanscient’s vision is to use generative AI to automate the whole process of designing multiphysics simulations.
“Our end goal is to provide a fully text-based interface where users can specify what kind of simulations they want to make, and the rest would then be automated,” says Aytekin. “It will take a lot of work to get there, but with Gemini we believe it can be done.”
Quanscient is a Finnish tech company specializing in cloud-based multiphysics simulation software for industrial processes. By making physical prototyping digital, it empowers researchers and engineers to accelerate hardware development, optimize designs, and predict material behavior with greater accuracy and efficiency across a range of industries.
Industry: Technology
Location: Finland
Products: Gemini, Gemini 1.5 Flash, Gemini 1.5 Pro, Vertex AI
