Prompting

Prompting is the practice of crafting text (prompts) sent to an LLM to guide its behavior. Effective prompting improves answer quality, reduces hallucination, and ensures the model follows constraints (e.g., cite sources, stay grounded in retrieved context).

Prompting in DocLD

• RAG chat — System prompts instruct the LLM to answer only from retrieved chunks and include citations
• Extraction — Schema instructions guide zero-shot extraction of structured data
• Reranking — Optional LLM-based reranking uses prompts to score chunk relevance

DocLD uses lower temperatures for factual tasks to reduce randomness and improve consistency.

Related Concepts

Prompting shapes RAG and extraction behavior. Schema instructions guide extraction; system prompts constrain hallucination and enforce citation use.