Cosmos Dataset Search is a semantic search platform for large video datasets, supporting text-to-video and video-to-video queries by generating embeddings with Cosmos-embed and running fast similarity search in Milvus (cuVS-accelerated).
An NVIDIA blueprint for building a video analytics agent that can summarize and answer questions over large video libraries by ingesting frames + audio, generating dense captions, and indexing the results for context-aware RAG.
A BioNeMo blueprint for designing peptide/protein binders from a target sequence using models like RFDiffusion, ProteinMPNN, and AlphaFold2 (including multimer), producing candidate binders and predicted complex structures in PDB format.
A BioNeMo blueprint for small-molecule discovery that combines GPU-accelerated protein structure prediction, generative molecule optimization, and docking (OpenFold2, GenMol, DiffDock) in a notebook-driven virtual screening workflow.
AI-Q is a “deep research” agent blueprint that generates structured reports by preferring internal documents via RAG, falling back to web search when needed, and iterating with a plan/review loop plus optional human-in-the-loop edits.
NVIDIA’s RAG reference stack: multimodal ingestion + hybrid retrieval + reranking + grounded generation, built from NIM microservices (NeMo Retriever + Nemotron) with deployable docker/k8s options.
An Omniverse digital-twin workflow for designing and stress-testing AI factory/data-center layouts—simulate power, cooling, and operations so you can iterate before hardware ships.
A production-grade “data flywheel” service built on NeMo microservices: turn real prompt/response logs into evals + LoRA fine-tunes, then surface cheaper models that still hit your quality targets.
Simulate, test, and optimize physical AI and robotic fleets at scale in industrial digital twins before real-world deployment.
Generate exponentially large amounts of synthetic motion trajectories for robot manipulation from just a few human demonstrations.
A 3D-first generative workflow: use real-time rendering as conditioning so you can iterate on backgrounds/lighting/styles without changing the hero product asset.
An Earth-2 starter app for weather analytics: multi-layer geospatial visualization plus AI forecasting, built around NVIDIA’s FourCastNet weather model.
An Omniverse blueprint for building interactive digital twins focused on fluid simulation and external aerodynamics-style workflows.