The MLOps Community fills the swiftly growing need to share real-world Machine Learning Operations best practices from engineers in the field.

MLOps Community is a collaborative platform for professionals working at the intersection of machine learning and operations, offering resources, events, and discussions on MLOps best practices and methodologies. By bringing together data scientists, engineers, and DevOps practitioners, MLOps Community enbles knowledge sharing and innovation in deploying, managing, and scaling machine learning models in production environments.

MLOps Community

DREAM is a blueprint for distributed RAG experimentation using Ray, LlamaIndex, Ragas, MLFlow, and MinIO on Kubernetes. It helps in comparing and contrasting different combinations of RAG parameters to find the best one. The architecture includes technologies like Ray, LlamaIndex, ragas, MinIO, MLflow, Jupyter notebooks, Kubernetes, and ArgoCD.

DREAM: Distributed RAG Experimentation Framework

<p>The thing that usually breaks these sweeps is combinatorics: chunk size x overlap x top-k x reranker is already hundreds of runs, and in my experience most of the variance comes from the chunking stage alone. Pinning the retriever and sweeping chunk boundaries first (semantic vs fixed, overlap ratio) narrows the grid a lot before you start burning GPU on generation configs. Also worth logging retrieval-only metrics separately from end-to-end answer quality - otherwise a good retriever hidden behind a weak prompt looks like a bad retriever.</p>
