Onnx

Fine-tuned MiniLM exported to ONNX and run inside an existing Spring Boot app. No vector DB, no Bedrock per query, no GPU. The post I needed when I started.