Generate and evaluate retrieval embeddings with FlagEmbedding
Use FlagEmbedding to choose BGE embedding or reranking models, encode documents and queries, evaluate retrieval quality, and feed stronger context into RAG workflows.
npx skills add agentskillexchange/skills --skill generate-and-evaluate-retrieval-embeddings-with-flagembedding
Invoke this workflow when an agent or RAG system needs a stronger retrieval layer before answers, search results, or memory recalls are trusted. The operator selects a BGE embedding or reranker model from FlagEmbedding, installs the Python package, encodes documents and queries, tests similarity or reranking behavior, and exports vectors or relevance scores to the surrounding retrieval stack. Reviewable inputs include the corpus, query set, selected model, chunking rules, batch settings, and evaluation questions. Reviewable outputs include embeddings, ranked results, retrieval metrics, and examples of improved or failed recall.
What this skill actually does
Use this instead of treating retrieval as a black-box vector database setting when the quality problem is model choice, reranking, multilingual coverage, multimodal search, or evaluation. The scope boundary is embedding and reranking preparation for retrieval pipelines. It is not a general model catalog or library listing; the skill is specifically about using FlagEmbedding’s documented BGE toolkit to generate, test, and hand off retrieval signals that another agent workflow can inspect.