Production RAG for a core banking suite
Banking clients needed faster, accurate answers from data spread across structured and unstructured sources, inside strict compliance boundaries.
I designed and built the retrieval architecture from scratch: ingestion, embedding, and vector search pipelines powering LLM-driven Q&A, with evaluation loops and audit trails.
Knowledge retrieval efficiency up ~35%; manual query resolution time down ~40%.
Stack: vector retrieval · embeddings · model-agnostic gateway · sovereign cloud