Shamrock Marketing: Technical knowledge RAG

Context

Tire retread and repair knowledge was spread across technical PDFs, standards, product specifications and workflow documents. A useful answer needed to fit the product and procedure in question. It also needed a path back to the relevant source for verification.

System and approach

Documents are divided into retrievable passages with workflow, product and standards metadata. Azure AI Search retrieves relevant context, and Azure OpenAI builds an answer from those passages. Document and page citations connect the response to the underlying technical material.

• Chunk standards, workflows, and product PDFs with metadata.
• Retrieve context through Azure AI Search.
• Use Azure OpenAI to synthesise answers from source passages.
• Return document and page citations for verification.

Delivered scope

The scope includes document ingestion, contextual retrieval, answer generation and source citations for technical knowledge lookup.

Technology

• Python
• Azure AI Search
• Azure OpenAI
• FastAPI
• Azure Functions
• PostgreSQL