volcengine-ai-search-rag

Implement retrieval-first answering with explicit indexing, retrieval, and grounding stages.

Execution Checklist

1. Confirm corpus source and chunking strategy.
2. Generate embeddings and build/update index.
3. Retrieve top-k context with filters.
4. Build grounded answer with citations to retrieved chunks.

Quality Rules

• Separate retrieval prompt from generation prompt.
• Keep chunk metadata (source, timestamp, id).
• Return confidence and fallback path if no hits.

References