Find and summarize relevant documents with AI
Even if the specifications are manageable, an ever-increasing mountain of relevant documents is piling up next to them. On the one hand, there are standards that need to be taken into account, and on the other hand, there are the applicable documents. This can quickly amount to hundreds or even thousands of pages of text.
Experts familiarize themselves with the relevant documents over many years and know the content well enough not to have to look everything up. New employees have a much harder time with this. Especially at the beginning, they ask the experts about every little thing, which means they have less time for other things.
Today it's about a topic of my own: I'm developing an AI system that solves this problem. Next week (15.7., 16:00) I will be organizing a Webinar in Englishweek after next (22.7., 16:00) in German. Here are a few details.
The problem
Through Increasing complexity and compliance requirements, the mountain of relevant documents is growing. This has been happening for decades. However, the growth has accelerated in recent years due to networking and the increased use of software.
The risks resulting from omissions have also increased: contractual penalties are comparatively harmless. It is worse when audits have to be repeated and this delays the market launch of a product. The worst-case scenario is a recall, or even Deaths caused by product defects.
AI can read documents
One possible solution is the use of Natural Language Processing (NLP). Here, an AI system analyzes the relevant documents. The challenge now is to support the user with this knowledge.
I am currently working on Semiantan AI-supported quality assistant that is supposed to do this. Semiant recognizes concepts in texts and their correlations. The cover image shows how such a system could support developers: The user must first define the context. The easiest way to do this is to select a requirement. The AI can then evaluate the individual chapters in the standards and relevant documents in terms of their relevance.
As individual chapters can be very long, Semiant creates a summary of 1-2 sentences for each relevant chapter. The list of chapters and summaries is then presented to the user. If required, the user can jump directly to the relevant chapter via a corresponding link.
Intermediate step Domain model / Glossary
The system described here is very ambitious. As an intermediate step, Semiant first generates a domain model from natural language texts. A domain model recognizes important concepts in texts and how they are related. A simple result that can be derived from a domain model is a glossary. This is already possible today, and I am looking for interested parties who would like to try it out.
Would you like to try Semiant? Then write me a short message.
Michael Jastram
Otherwise, I will also publish the solutions next week and the week after in Webinars as described at the beginning.
