Multi-hop reasoning is an AI system’s ability to answer a question through several successive inferences, often combining information from different sources. Example: “Which department is headed by the person responsible for project X?” first requires identifying the responsible person and then their department. Such questions are demanding for search and RAG systems because every intermediate step must be found and linked correctly.
Retrieval-augmented generation combines search with text generation: before answering, the system searches a knowledge collection – such as manuals, policies or FAQs – for relevant passages and passes them to the language model as context. The answer is therefore based on current, company-specific content rather than only on the model's training knowledge. This reduces invented statements, makes sources traceable and allows updates without retraining.
Question answering (QA) is a field of natural language processing in which a system answers questions posed in everyday language directly, rather than returning a list of search results. Modern systems look for relevant passages in a knowledge base and compose an answer from them, ideally with a source reference. Typical uses are customer service, internal knowledge bases and manuals – for instance asking the staff handbook ‘How many days’ leave do I get in my first year?’
LazyGraphRAG is an approach to graph-based retrieval-augmented generation presented by Microsoft Research in 2024. Unlike GraphRAG, it does not have a language model create summaries during indexing; instead, it identifies concepts and their co-occurrences using classic natural language processing, and the language model is only used at query time. According to Microsoft, this brings indexing costs down to the level of basic vector RAG. It is particularly suited to questions that require an overview of large document collections.
Model chaining means connecting several AI models or AI steps in sequence so that the output of one step becomes the input of the next. Complex tasks can thus be broken down into manageable sub-tasks that can be checked and improved individually. Example: one step summarises a customer enquiry, a second assigns it to a department and a third drafts the reply.
