# How does NVIDIA AI-Q Blueprint improve search?

Source: https://zilliz.com/ai-faq/how-does-nvidia-aiq-blueprint-improve-search
Author: zilliz

NVIDIA AI-Q Blueprint is an open-source reference implementation that fundamentally improves enterprise search through agentic reasoning and multi-phase research. Traditional search (keyword or semantic) returns documents; agentic search reasons over documents, synthesizes findings, evaluates reliability, and presents structured answers with citations. AI-Q implements a two-tier architecture: shallow research agents for quick answers (bounded to 10 LLM turns), and deep research agents for comprehensive investigations of complex topics.

AI-Q's hybrid model strategy improves both quality and cost: frontier models (GPT-4 level) handle orchestration decisions and synthesis, while NVIDIA Nemotron open models perform research tasks—cutting query costs by over 50% compared to frontier-only approaches while maintaining top accuracy. The blueprint tops the DeepResearch Bench accuracy leaderboard, the gold standard for evaluating research agent quality.

Key improvements to search: (1) **Agentic reasoning** over results rather than static ranking, (2) **Multi-phase research** that breaks complex queries into subtasks, (3) **Iterative refinement** as agents discover knowledge gaps and perform additional research, (4) **Citations and provenance** showing which sources support each claim, (5) **Cost-optimized inference** through hybrid model selection. Integration with [Zilliz Cloud](https://zilliz.com/cloud) provides fully-managed vector database infrastructure supporting large-scale RAG for research agents. Queries against enterprise knowledge bases return relevant documents, which agents synthesize into trusted business insights—transforming search from retrieval to reasoning.