About This Session
You can compare AI to a calculator: just because a tool produces results doesn’t mean the underlying science has disappeared. Challenges that have existed in computer science for years are simply being addressed today using different methods.
This is well illustrated by the example of entity extraction. Let’s imagine a Large Language Model (LLM) analyzing thousands of medical studies and automatically extracting relevant information such as medications, diseases, and side effects. The goal is to extract structured data from unstructured text—for instance, to efficiently evaluate correlations or interactions.
To achieve this, an LLM must also handle ambiguities (e.g., “Java” as a programming language, an island, or coffee), take context into account, and correctly recognize nested entities. These challenges persist regardless of whether rule-based methods, classical machine learning approaches, or modern LLMs are used.
Using a specific use case, this presentation demonstrates that while we now have more powerful tools at our disposal, classic challenges in computer science persist. Building on this, a practical architecture is presented: LLMs for semantic extraction, vector databases for contextualization, and n8n for orchestrating robust data pipelines. Step by step, we’ll show how a naive prompt can be transformed into a stable, traceable, and extensible solution.