Where was that picture again...? Ask the AI!

Do you know this?

I have to make a presentation for a customer about the possibilities of acoustic calculation of noise emission from gearboxes. I vaguely know that I did acoustic calculations on a MAN engine many years ago, but I've been out of the subject for a long time and many of my engineers have calculated many projects on this or similar topics in the meantime without me even realizing it. Perhaps one or two of the engineers no longer work for us and it was before Maik's time, who might still have the best overview here.

Now the search is on, who did what, when and for whom. Are there any pictures of old projects that fit the bill and are old enough that we can show them without any problems?

Are there any offers and reports on this topic that would make my work easier? And if so, which ones?

Valuable information that is buried somewhere and that we would like to use for current projects, if we could find it.

An initial approach a year ago with a start-up was not useful, as the answers were far too unspecific and didn't really help us.

While shopping in the supermarket, I bumped into a former acquaintance with whom I had worked together at the Technology Center in Heidenheim around 35 years ago. We had each founded our own startup back then, but it used to be called something else. Back then, he developed software and I bought software to solve engineering problems because programming wasn't really my thing. Since then, he has worked as a freelancer for large companies and has written a wide variety of programs and, in this context, has also dealt extensively with the possibilities of AI in recent years, e.g. in a project on autonomous driving, where he worked as a developer. He was able to explain to me in simple terms what language models are, how they work and what open source software is available that you can do a lot with.

I then told him about my image search woes and just two days later he wrote me a small bot as a prototype or proof-of-concept solution that fished out the right images for me from the archived reports.

Not a great user interface, but exactly what I was looking for.

When searching for an image of a flow simulation for a lamp, I got the following result:

That doesn't sound bad, and nowhere in the text of the report was it mentioned that it was a low-beam bulb.

Well, that's when the idea was born that what he was developing for us might also be of interest to you. As a freelancer, he doesn't have the reach to publicize the topic, so we decided to see if there was a market here for a joint company.

I then asked him to put the whole thing together for me to use in a blog article and voilà, there's the advertising block:

We face the daily challenge of documenting complex projects and efficiently accessing previous work. Our quotations and project reports contain valuable information - but until now they were mostly available as unstructured PDFs. This made it difficult to search for specific content or quickly evaluate past projects. That's why we implemented a customized AI solution to help us make our work even more efficient.

Over the years, we have accumulated numerous offer documents that provide valuable insights into our work. However, these were available in continuous text form, meaning that targeted analysis or reuse would have been extremely time-consuming. In addition, our project reports contain numerous illustrations that provide crucial information but could not be easily captured by a text search.

To overcome this challenge, we have developed an AI application that supports us in analyzing and structuring our documents. We rely on two powerful functions:

A locally operated open source language model analyzes our unstructured offer PDFs and converts them into structured data records for an SQL database. This allows us to search and evaluate our offers in a targeted manner - without having to enter content manually.

We also store the quotes in a vector database to enable semantic searches. This means that we can simply enter a question in natural language - for example: “What offers have we created for automotive suppliers in the last two years?” - and the AI immediately delivers the relevant results. Depending on the query, a system of autonomous AI agents decides whether a classic SQL query or a semantic search is the best method.

Our project reports often contain valuable diagrams, illustrations and plans. Until now, however, these images were difficult to find as they were not directly searchable.

With our new application, we automatically extract these images from the PDFs and analyze them using AI. This generates linguistic descriptions of the images, which in turn serve as the basis for a semantic search. For example, we can search for a specific chart by simply entering a description - and the AI finds the relevant images, even if they are stored in completely different reports.

Thanks to these AI technologies, we not only save countless hours of manual work, but also gain completely new ways of using information. Our engineers can access relevant content faster, process recurring requests more efficiently and make better use of their previous work.

Another key advantage is that everything runs locally on our own systems - which means 100% control over our data and maximum security.

This innovative solution was developed in collaboration with an AI and IT solutions provider. This provider has extensive experience in implementing customized AI applications, including AI-powered solutions for customer support and other business processes. By working together, we were able to ensure that our solution was perfectly tailored to our needs.

Perhaps you are facing similar challenges in your company. Perhaps you too have a large number of documents that are full of valuable information - but are difficult to access. The good news is that AI can help, and it can be tailored to your individual needs.

If you would like to find out more about how AI can optimize your processes, please get in touch with us. Together we'll find out how we can make your data more intelligently usable!

So much for the advertising blog. Doesn't sound as smooth as my own texts, but AI is much younger than I am 😊.

I think engineers should talk to IT nerds much more often. That's why my friend and I have arranged a regular jour fix. I don't want the AI to rationalize my engineering job without me noticing.

So here is my offer to you: If you are interested in using similar software, please get in touch with me.

PS: He actually claimed that in a few years most of our work as engineers will be done by AI. I don't think it will be quite so bad if we learn to use AI efficiently in the right places. And as long as AI can't make coffee...😉