Due to the high complexity of internal knowledge regarding production and administrative processes, the customer aims to simplify access to knowledge for all employees, regardless of their experience or length of service. For this purpose, PTA is designing and developing a prototype of an AI system that can answer questions based on the company's documentation. The system is intended to be easy to use and capable of flexibly answering questions in natural language without requiring prior knowledge or strict formatting guidelines. Since company documentation represents trade secrets and proprietary know-how, a local operating environment is targeted, running entirely on the company's own server hardware without any external communication over the internet. To sustainably optimize operating costs and maximize the benefits of AI deployment, the system is being developed based on a resource-efficient RAG architecture with the integration of medium-sized open-source language models.
Supplement
The provided company documentation comprises a total of 150 work instructions, each of which must be analyzed individually in order to derive relevant use cases in the form of a chunking strategy as well as test questions. The embedding of the company documentation into the knowledge base of the AI system is carried out using locally operated open-source software solutions that do not require internet access for data processing or storage. After downloading the selected LLMs, no internet communication is required for generating responses through GenAI either, which ensures confidential and autonomous functionality. The RAG architecture enables flexible adaptation and extension of individual components according to customer requirements without the need for extensive fine-tuning of the language models. Finally, the basic AI behavior can already be controlled in the prototype through system instructions and operational configuration in order to increase trustworthiness.
Subject description
The prototypical system development enables the customer to gain the earliest possible practical insights into the functionality as well as impressions of the benefits of AI deployment in their own use case. This facilitates strategic decision-making regarding the functional scope and investment requirements of the planned system. Furthermore, all relevant stakeholder groups of the customer are given the opportunity to directly experience the potential future approach to democratizing the company's knowledge. The feedback and optimization suggestions gathered during the prototypical phase provide a solid data basis for developing and implementing the future production system in a targeted, trustworthy manner, while reducing both financial and time expenditures.