AI Support Assistant (RAG)

A course business was spending significant time answering repetitive questions already covered in its own documentation - support that did not scale with the audience.

An AI assistant that indexes all of a business's documents into a vector database, then answers questions on demand by retrieving the most relevant passages and generating accurate, source-grounded replies through a chat interface.

Making a retrieval-augmented AI system economical and reliable. I built a seven-layer cost-optimisation strategy - deduplication, batching, multi-tier caching and strict token budgeting - that cuts AI API spend by an estimated 70%, plus consistency handling across the vector and relational stores so failed indexing never leaves orphaned data.

Per-query cost held to a fraction of a cent. Handles unlimited concurrent users around the clock. Conversation memory persists across sessions for natural follow-up.