Accelerate post-training of end-to-end autonomous vehicle stacks with vector search and retrieval for large video datasets.