Event details
You have options across Azure to accomplish report summarization. Within the context of RAG and Azure AI Search, you can use prompt engineering techniques to accomplish your desired outcome by giving the LLM specific instructions to summarize retrieved documents. Relevant documentation on prompt engineering for Azure OpenAI is here. Within Azure AI Services Language Service there is also an API with native document support for summarization (please see this announcement). The right detailed steps are going to be dependent on exactly what you are trying to accomplish in terms of summarization (extractive, generative, etc.) and where you are starting from (already have the report document, must retrieve it first, etc.).
- Our users will already have the PDF/DOCX/etc. file containing the report for which they want a summary. - Not sure I know LLM enough to yet answer "extractive, generative, etc.", but imagine a 20-30 page report about which you want a 3 paragraph summary. - What specifically are "retrieved documents" where you write "giving the LLM specific instructions to summarize retrieved documents"?
Retrieval Augmentation Generation (RAG) is the focus of this AMA and is an architecture that augments the capabilities of a Large Language Model (LLM) like ChatGPT by adding an information retrieval system that provides grounding data. Put differently, it is the combination of an LLM (large language model) that has improved knowledge through its ability to leverage a retrieval system to pull in relevant information not contained within its training data. In that context a retrieved document is a search result that is handed to the LLM that contains information that the LLM will use to compose a response. The LLM could be prompted to summarize the document as desired.
That said, you are noting that you already have the document and therefore don't need to perform the retrieval step but want to jump directly to summarization. Among other options within Azure you canleverage the Language AI Service's summarization capabilities or to use the Azure Open AI Playground to perform summarization. Both links will include more information about your options and can help you determine what is best for your use case.
We'll tag the Azure AI Services Language team for additional information that can aid you.
