A financial institution is developing an internal system to help employees quickly find specific information within a vast repository of financial documents, including PDFs, Word files, and scanned images. The system needs to allow natural language queries and provide precise answers, not just document links. The institution also wants to ensure that the system can handle new, complex financial terminology that is not commonly found in general language models. Which Azure AI approach should they prioritize?
- ADevelop a Custom Text Classification model to categorize documents and link to external knowledge bases.
- BUtilize Azure AI Language's built-in Text Summarization and Key Phrase Extraction.
- CImplement Azure AI Search with a custom-trained LUIS model for query understanding.
- DDeploy Azure OpenAI Service with a Retrieval Augmented Generation (RAG) pattern, integrating Azure AI Search.
Show answer & explanationAnswer & explanation
Correct answer: D. Deploy Azure OpenAI Service with a Retrieval Augmented Generation (RAG) pattern, integrating Azure AI Search.
The requirement for 'natural language queries' and 'precise answers, not just document links' points to a generative AI solution. The need to handle 'new, complex financial terminology' and 'vast repository' suggests combining a powerful language model with a robust search capability that can index custom data. The RAG pattern with Azure OpenAI Service and Azure AI Search is ideal for this, as it grounds the generative model's responses in the specific, proprietary data, ensuring accuracy and relevance while handling specialized terminology.
Why the other options are wrong
- A. Custom Text Classification categorizes documents, but doesn't provide precise answers to natural language queries or handle new terminology in a generative way.
- B. Summarization and Key Phrase Extraction are analytical tools, not designed for natural language querying and precise answer generation from a large corpus.
- C. LUIS for query understanding combined with AI Search would provide document links but struggle with precise answer generation and complex terminology without a generative model.
Retrieval Augmented Generation (RAG)
A generative AI pattern that combines a large language model (LLM) with a retrieval system (like a search engine) to enhance the LLM's responses by grounding them in specific, up-to-date, or proprietary data.
- LLM generates answers, but based on retrieved context.
- Reduces hallucinations and provides more accurate, relevant responses.
- Ideal for enterprise search, Q&A, and chatbots needing specific data.
Memory trick: RAG retrieves and Generates Accurate Grounded answers.