ibm-self-serve-assets/SuperKnowa

Built on IBM's watsonx, the framework provides production-ready RAG pipelines with pluggable components across the full stack—document indexing (Elasticsearch, Solr, Watson Discovery), neural retrieval, re-ranking, LLM in-context learning, and fine-tuning (QLORA for Falcon/LLAMA2). It includes integrated evaluation tooling (BLEU, ROUGE, BERT scores) via MLflow for experiment tracking and leaderboards, plus an AI Alignment Tool to capture human feedback and measure model helpfulness and accuracy. Configuration is YAML-driven, allowing rapid assembly of production pipelines validated at scale across knowledge bases from 1M to 200M documents.

116 stars. No commits in the last 6 months.