Retrieval-Augmented Azure OpenAI Assistant

Overview

Designed an end-to-end Retrieval-Augmented Generation pipeline on Azure OpenAI that lets staff query products, policies and procedures in plain language instead of searching portals by hand. Consolidated and cleaned semi-structured and unstructured content from several internal sources, standardising formats, removing duplicates and aligning metadata into a single reliable source of truth.

Implemented document chunking with tuned overlap and generated vector embeddings, stored in Azure Blob Storage for low-latency, context-aware retrieval, then connected the vector store to Azure OpenAI behind a chat interface. Worked with operations and support stakeholders to iterate prompts, chunking strategy and retrieval settings, improving answer relevance and reducing follow-up questions. Instrumented usage analytics and feedback logging to track adoption and answer quality, which surfaced common information gaps and guided content curation.