Cutting Engineers Tedious Task Time to Seconds with Agentic AI Workflow Automation

Synera has built, with Vstorm’s applied expertise, a text-to-workflow AI agent to flip the core concept of engineering modelling, optimization and prototyping upside-down.

According to a McKinsey report, the manufacturing industry is still reluctant to use AI agents in their workflows and operations, with 91% of organizations stating that they are not using them at all. Synera, a German-based company, aims to challenge this shortcoming with Vstorm’s support implementing AI-powered workflow optimization.

Vstorm’s impact, the TL;DR:

• Vstorm built a text-to-workflow system for Synera’s low-code platform and AI Agent environment using LLMs, RAG, and validators
• The system operates using graphical nodes inside of Synera’s visual engineering-automation platform
• The node-based operations initiated by text input was possible thanks to Synera’s own interpreter that converts nodes-to-pseudo-python and back
• The Synera-Vstorm teams managed to overcome issue of not having a dataset of examples to build on by creating a set of prompt-code-workflow triads from scratch
• The automation accelerates the workflow creation that can be tweaked by users to ensure that it fits to their needs

The client

Synera operates an engineering support low-code platform and AI Agent environment which integrates with popular CAD software and solutions. Their agents and automations accelerate the product development process by up to 10x times, mostly with the reduction of workflow complexity and cost control. The platform enables over 150 workflow automations for companies and organizations like NASA, Airbus, BMW, Hyundai and Henkel, among others.

Synera was founded in 2018 in Bremen, Germany, and supports integrations with the leading providers of CAD tools, including Altair, Autodesk, Hexagon, PTC and Siemens.

“Our product can save years of development time for our customers from automotive and aerospace sectors.” – Andrew Sartorelli, Head of Product Management at Synera

The challenge

Synera’s platform is versatile and powerful, letting engineers and specialists build automations and smooth their processes. But this versatility comes with a price.

Automations are built in an editor that resembles a graph with blocks, each representing unique operations, that can be connected to communicate or sequence operations. Contrary to multiple other low-code tools, this one is also fairly low-level. With that, blocks can be used in a myriad of combinations delivering a plethora of results. But on the other hand, it takes time and a deeper understanding of the system to build more complex workflows, creating barriers and requiring time investment that could have be spent in a different way, for example, in designing machines.

To solve that, Synera sought AI-powered automations to apply to their platform in hopes to make the process smoother and boost the user experience.

Our work

Vstorm started the cooperation by gaining a deeper understanding of the nature of the problem to solve. Since Synera’s team has been exploring the concept of supporting workflow creation in their app with AI to cut engineering time, that effort served as a springboard for the Vstorm team.

The core idea

Synera decided to introduce a system that would enable users to interact with a chat interface to deliver desired complex node-based structures to users’ liking, basically a text-to-workflow model to garner the full yield of workflow automation benefits. The goal was ambitious and the process could be seen as intimidating at first.

“A lot of companies have this strategic vision of AI, but how you bring it down to reality is really a challenge.” – Andrew Sartorelli

Early in the project, Vstorm tested multiple methods to create a comprehensive report alongside practical recommendations on how to achieve the desired outcome.

“We had our own ideas, but the Vstorm team took the projects and visions for two or three weeks and came back to us with a comprehensive report on what is possible and what is not.”” – adds Andrew

As the path forward crystallized, we began developing the solution.

From the very beginning, the core of the solution was PydatnicAI framework used as an environment to develop agents and experiment with multiple available solutions.

We experimented extensively with different models and architectures to find the best approach. PydanticAI provided a stable, reliable environment that enabled us to test multiple workflows and compare results.

“The VStorm team could easily swap components or remodel the system within the same framework, accelerating the delivery process by 40% without sacrificing reliability.”

• comments Andrew.

How the language converts to nodes (and back)

To make text-to-workflow possible, Synera team first created an interpreter to describe nodes and node-based structures using pseudo-python code. The existence of such code made it possible to manipulate nodes with code using LLMs and convert the code back to nodes. Such code is legible for an LLM, so the system can operate using it, making it possible for humans to express intent (in plain language) that is then converted to code, which converts again into nodes seen in the software panel.

Building dataset from scratch

Data scarcity is a challenge in nearly every Artificial Intelligence project, the data may be badly annotated, datasets can be biased, or there may be not enough data to train the solution. In this particular case, there was no existing data at all. But there was a great asset to make use of, a library of thousands of sample automations owned by client. So the idea was that this repository of samples could be turned into a dataset for an LLM to use.

And so, in order to make the system generate workflows using prompts, the process had to be reversed first into a workflow-to-text manner. With that, a viable database was complete, and our team could start building the solution. The dataset was also enriched with completely synthetic data generated by the LLM, so the amount of sample data to learn on was substantial enough to deliver a robust solution.

Making the model work

The team prepared a dataset consisting of prompt, code, and workflow triads, enabling the model to connect each to the other and spot the patterns necessary to deliver text to the workflow builder.

The model works in the other way around to a dataset builder. The Large Language Model inside analyzes the prompt, which is later turned into pseudo-python code representing the user’s demand. Later, the interpreter has to turn the code into the desired workflow, represented as a graph.

An unexpected challenge in this stage came from the diligence of modern LLMs. In some cases, the pseudo-python generated by the LLM was closer to the full version of the programming language, adding functionality which is missing in the pseudo code. This was caused by LLM’s tendency to deliver “as good a job as possible,” where the model delivered better solutions from the programming logic, yet ill-fitted to the required context.

This challenge required the building of a validation mechanism which checks the code before sending it to the interpreter. As a result, the code is sent back and forth between the LLM and validator until working pseudo-python is produced.

RAG system support for less work

With a large database of prompt-code-workflow triads, it became necessary to make the solution more stable and reliable. So our team implemented a RAG-based solution to support the model’s work.

Based on the user’s prompt, RAG searches the database of triads mentioned above, looking for ones as similar as possible. When spotting them, RAG delivers the top three picks to the LLM as “inspiration,” reducing both the chance of a mishap during code generation and the likelihood of misinterpreting the intent of the user.

RAG was also helpful due to the documentation already existing in some workflows, so the system gets an even deeper understanding of what has to be done and why. The system also processes the documentation of the nodes to use in the workflow designer, delivering a necessary context for the LLM to use.

Testing and validation

Last but not least, the system has to be extensively tested and validated in search of mistakes and mishaps. With several components that have to seamlessly operate, the LLM, the interpreter, and the RAG system, there are many points where something could go wrong. That is why the solution is being rolled out gradually, along with tests done by in-house experts and hired specialists alike.

“After three months into the POC we started to see really the ability to generate complex workflows from a simple text prompt.” – comments Andrew

Vstorm’s solution

With Vstorm’s intelligent process automation, Synera can now deliver a product that fits its founders’ vision. The system’s goal is to turn each prompt into workflow automation tools which fit as close as possible to the user’s intent, which can then be fine-tuned and adjusted to the explicit needs, saving hours of work.

Thereby it not only liberates the time required to work, but also encourages engineers to tinker around and test if some other aspects of their work can be automated or streamlined.

“With our current version of text-to-workflow [agent] it takes about 2 minutes to write a text prompt and I can create a workflow that would take me up to an hour to put together with a lot of manual clicks.” – concludes Andrew Sartorelli