Founded in 1911 and headquartered in the Netherlands, Kendrion (AMS:KENDR) is one of the world’s largest manufacturers of electromagnetic systems for a wide range of industries, from high-end automobile companies to robotics manufacturers.

• Develop an AI-powered co-pilot for managers and senior executives, enabling seamless exploration and reasoning through decades of historical project data.
• Implement multi-layered safety mechanisms to minimize hallucinations in AI-generated responses.
• Build a robust, dynamic, embedding-driven knowledge base supported by a data pipeline to extract, preprocess, and vectorize project documents as they are updated.

We designed and implemented an AI co-pilot powered by large language models (LLMs) which connects directly to Kendrion’s extensive project data repository, allowing users to retrieve precise and relevant insights effortlessly.

Core Components:

• Dynamic Knowledge Base: The AI leverages a dynamic knowledge base where project documents are continuously updated, preprocessed, and vectorized. This ensures that the most up-to-date and relevant information is always accessible.
• Contextual Search Engine: To enhance retrieval efficiency, we implemented a semantic search engine that uses numerical representations (embeddings) of natural language to match user queries with the most pertinent documents.
• Answer Generation: The LLM synthesizes retrieved context into coherent, precise answers, optimizing both token usage and computational efficiency. This hybrid approach minimizes costs while maintaining high accuracy.

We used a combination of semantic and syntactic search algorithms to match user queries with the most relevant data points. By focusing on contextually rich semantics, it filters irrelevant content and delivers precise responses tailored to user needs. Additionally, users can refine search results by selecting specific projects to focus on, allowing for granular control over the retrieval process.