AgentSkillsCN

rag-ops

RAG管道运维。用于检查Qdrant集合、调试数据摄入过程，以及优化检索参数。

SKILL.md
--- frontmatter
name: rag-ops
description: RAG pipeline operations. Use for inspecting Qdrant collections, debugging ingestion, and optimizing retrieval parameters.

RAG Operations Specialist

This skill assists with the management and optimization of the RAG (Retrieval-Augmented Generation) pipeline.

Capabilities

1. Inspect Vector Database

Use the bundled script to check the status of your Qdrant collection and peek at indexed documents.

bash
python skills/rag-ops/rag-ops/scripts/inspect_collection.py

2. Ingestion Debugging

When files fail to process or retrieval is poor:

  • •Check backend/ingest.py for chunk_size and chunk_overlap.
  • •Standard recommendation: Start with 1000/100. If answers are missing context, increase overlap to 200.
  • •Verify backend/data/files.json aligns with Qdrant contents.

3. Retrieval Tuning

To improve chat quality:

  • •Adjust k (number of retrieved docs) in backend/rag.py.
  • •Current model: nomic-embed-text (via Ollama).
  • •If irrelevant context is retrieved, consider re-ranking or stricter similarity thresholds.