Automate PLC Ladder Programming
with Generative AI
Leave routine ladder programming to AI —
it reads specifications, drawings, and more.
Common Challenges in Control Engineering
With AI programmer "Ladder-kun",
ladder programming is automated! (*)
* Currently in beta. Some types of programs are not yet supported.
Expected Benefits
What Makes Ladder-kun Different
Point 1
A proprietary, GENIAC-backed ladder LLM
General-purpose LLMs are trained mostly on web data and struggle with ladder programs, which have little presence online. Airion fine-tuned its own ladder-specialized LLM through "GENIAC," a Japan-made foundation model program led by METI and NEDO. On benchmarks built from equipment used in real operations, our model outperformed general-purpose LLMs by 40.0 pp in compiler pass rate (35.4% → 75.4%).
Point 2
Support for company-specific coding standards
To overcome the probabilistic nature of LLMs, we first assemble pre-registered templates via RAG to guarantee company-specific coding rules, then use the LLM only for the rest. Development works regardless of your PLC vendor.
Point 3
Proven on the factory floor
In a PoC with equipment manufacturer Nagano Automation, we cut ladder programming hours by approximately 20%, with further gains expected as accuracy improves. We also tailor the AI to each customer's specifications and coding style as needed.
Implementation Flow
FAQ (Security & Reliability)
Q1. Could the content of my ladder programs leak or be misused?
The data you provide is your valuable technical information, and we manage it strictly.
Data is processed within the secure AWS (Amazon Web Services) cloud environment, with access restricted at the network level. Even within Airion, data access is limited by role — it is never freely visible to everyone.
Q2. Will my data be used to train AI models?
No.
The AI models we use (Anthropic Claude, OpenAI GPT-4o, etc.) are contracted as commercial APIs. Each provider's terms of service explicitly state that input data is not used for training. No special configuration is required on your side.
Q3. Is there a risk in entering personal or confidential information?
The service fundamentally processes files that you specify. In addition, you may optionally enter free-form instructions (prompts) to assist processing.
Ladder programs contain technical information about equipment and control logic, but since this information is necessary for processing, you can enter it as is. The system is not designed in a way that allows input data to leak externally, so you can use it with confidence.
Note that ladder program comment fields sometimes contain the names of past engineers. Because program content itself often constitutes your technical know-how, we treat all input files as confidential information.
Q4. Which AI models do you use?
We use models from leading, trusted AI companies such as Anthropic's Claude and OpenAI's GPT-4o.
Claude is used via Amazon Web Services (AWS) Bedrock, running on AWS's stable infrastructure for reliable operation.
Q5. In which country is my data processed?
For models accessed via AWS Bedrock, the data processing region can be specified at contract time. If your internal policies require a specific processing region, please feel free to consult us.