<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> The NVIDIA AI Research Assistant blueprint enables on-premise deep research report generation. Targeting research analysts and developers, it leverages internal data and web search to produce detailed reports, enhancing research efficiency and depth.

How It Works

This blueprint uses an agent for deep research: planning, parallel searching, writing, reflection, and source citation. It queries NVIDIA RAG for multimodal documents and optionally Tavily for web search, prioritizing internal data via an LLM-as-a-judge. The architecture includes a demo frontend, backend RESTful API (aiq-aira Python package), and Nginx proxy, built on NVIDIA NeMo Agent Toolkit and Microservices.

Quick Start & Requirements

Deployment uses Docker Compose or Helm, guided by a "Get Started Notebook." Prerequisites include Ubuntu 22.04, 250 GB disk, NVIDIA Container Toolkit, GPU drivers (>= 530.30.02), and CUDA (>= 12.6). Significant GPU resources are mandatory, from single high-end GPUs to multiple (e.g., 4x H100 80GB for the full blueprint). An NVIDIA AI Enterprise license, NGC API keys, and an optional Tavily API key are required.

Highlighted Details

• Comprehensive deep research workflow: planning, parallel searching, writing, reflection, and source citation.
• Parallel search prioritizes internal multimodal documents queried via NVIDIA RAG.
• Human-in-the-loop capabilities for interactive report refinement and Q&A.
• Integration with NVIDIA RAG for querying text, charts, and tables.

Maintenance & Community

The provided README does not detail specific maintenance contributors, community channels, or sponsorship information.

Licensing & Compatibility

The blueprint software is governed by Apache License 2.0. Models are subject to NVIDIA Community Model License, NVIDIA Open Model License Agreement (Llama-3.3-Nemotron-Super-49B-v1), and Llama Community Licenses. NeMo Retriever extraction is Apache-2.0. Compatibility requires an NVIDIA AI Enterprise license for local NIM hosting. Users must review all third-party licenses.

Limitations & Caveats

Basic prompt filtering does not prevent SQL injection, code execution, or XSS; end-users must implement production security (AuthN/AuthZ, logging, vulnerability management). The blueprint is "as is," with users responsible for container security, updates, and OSS package currency. Deployment demands substantial, high-end GPU hardware.