Build a Deep Research Agent
NVIDIA Deep Learning Institute
Last updated June 18, 2026
Build a Deep Research Agent is a four-hour, hands-on workshop from NVIDIA's Deep Learning Institute for developers who want to assemble an autonomous research system rather than a simple question-answering bot. Working in Jupyter notebooks, you deploy NVIDIA's Nemotron reasoning models, stand up a retrieval-augmented generation (RAG) pipeline over a vector database, and wire in web search so the agent can plan a research task, gather evidence from private documents and the public web, and write up a cited report. The work is built around NVIDIA's AI-Q Research Assistant Blueprint and emphasizes running real services end to end, with human approval steps along the way. The labs call NVIDIA-hosted model endpoints and Tavily web search, so you generate two free API keys (NVIDIA NGC and Tavily) as a first step.
What you'll learn
- Deploying NVIDIA Nemotron reasoning models via hosted endpoints
- Building a RAG pipeline with a vector database and multimodal document ingestion
- Orchestrating agentic workflows with planning and human-in-the-loop approval
- Running parallel searches across private documents and the web to generate cited reports
Frequently asked questions about Build a Deep Research Agent
Who is Build a Deep Research Agent for?
Developers with basic Python and some AI/ML familiarity who want to build an autonomous, citation-producing research agent.
Is Build a Deep Research Agent free?
No — Build a Deep Research Agent is a paid course.
What are the prerequisites for Build a Deep Research Agent?
Basic Python and familiarity with Jupyter notebooks; understanding of AI/ML concepts (LLMs, embeddings) is helpful, and familiarity with Docker is beneficial. The labs require free NVIDIA NGC and Tavily API keys, with instructions provided in the lab.
Do you need to code for Build a Deep Research Agent?
Yes — Build a Deep Research Agent involves hands-on coding.
Why we suggest this course
For developers who already grasp RAG basics and want to go further — orchestrating reasoning, retrieval, and web search into a single agent that produces sourced, multi-step reports. The distinct takeaway is hands-on experience with reasoning models and human-in-the-loop research workflows on a production-style blueprint.