Owned end-to-end quality validation for a production RAG pipeline and a fine-tuned Qwen 2.5 7B model powering an enterprise knowledge assistant. Defined acceptance criteria and built a golden test set so retrieval quality was measured objectively rather than spot-checked. Verified 94.2% retrieval similarity and 1.5s response latency, 40% better than baseline. Also built integration and performance suites for the Flutter and FastAPI application layer, holding 99.7% uptime across releases.