AI & ML

Launch distributed Megatron-LM training jobs on a SLURM cluster with a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules, container conventions, monitoring, and per-rank failure diagnosis.

Plan, configure, and chain Nemotron model customization steps into single-step or multi-step pipelines for curation, translation, fine-tuning, RL alignment, benchmarking, checkpoint conversion, optimization, and evaluation.

Trains vision transformers via self-distillation without labels for self-supervised visual representation learning, and supports export and inference of NVDINOv2 backbones.

Trains, evaluates, exports, quantizes, and runs inference for Sparse4D multi-camera temporal 3D object detection and tracking models using TAO.

Selects, configures, deploys, verifies, debugs, or tears down a VSS profile (base, search, lvs, warehouse, edge) for NVIDIA's video search and summarization stack.

Search archived video using natural language, ingest video files or RTSP streams, and manage ingested sources.

Guides AI coding assistants to the official NemoClaw documentation via MCP server or Markdown files for installation, configuration, operation, and troubleshooting.

Collects launch inputs and runs preflight checks before executing TAO workflows such as AutoML, training, evaluation, inference, export, TensorRT engine generation, or DEFT jobs on supported platforms.

Embeds target and source image parquets, then mines nearest-neighbour source images for augmentation in VCN AOI workflows.

Trains, evaluates, and runs inference for BEVFusion multi-sensor 3D object detection models that fuse LiDAR and camera data in bird's-eye-view space for autonomous driving.

Train, evaluate, export, quantize, and run inference for a TAO OneFormer model that performs panoptic, instance, and semantic segmentation using task-conditioned queries.

Trains, evaluates, exports, quantizes, and runs inference for SegFormer semantic segmentation models using NVIDIA TAO.

Queries video analytics incidents, alerts, metrics, and sensor data from Elasticsearch via the VA-MCP server.

Summarize recorded video clips using the LVS microservice with a VLM fallback, producing a narrative summary with timestamped events.

Select, validate, patch, and deploy existing NVIDIA Dynamo Kubernetes recipes for model serving with GPU support.

Guides installing Earth2Studio via uv or pip, selecting model extras, and configuring environment variables.

Benchmark Jetson LLM/VLM serving performance across vLLM, llama.cpp, and Ollama with structured JSON output.

Generates synthetic CT volumes and masks using NVIDIA's rectified-flow pipeline for medical imaging research.

Generates synthetic brain MRI volumes using NVIDIA's NV-Generate-CTMR workflow, with configurable modality and random seed.

Navigate the PhysicsNeMo repository by discovering model families, datapipes, and examples through live file search, without writing training code.

Generates or updates a governance skill card for an existing agent skill directory by analyzing source signals, building a grounded JSON context, and rendering a deterministic markdown card.

Train, evaluate, export, and run inference for CenterPose models used in 6-DoF object pose estimation with keypoint regression.

Detects generalizable learnings from problem-solving interactions and proposes skill updates to improve future performance.

Start or patch Dynamo router modes and run router endpoint smoke checks for round-robin, KV-aware, least-loaded, or device-aware routing.

Find Earth2Studio models, data sources, and examples for weather/climate use cases by consulting live documentation and verifying compatibility via the lexicon system.

Run evaluation tasks against a NeMo Platform server using the Evaluator plugin CLI and Python SDK.

Guides agents through the full lifecycle of NeMo-RL experiments: understanding recipes, launching reproducible runs, analyzing results, and preserving human oversight with git and TSV logs.

Lists TAO Skill Bank capabilities, models, and AutoML support by running packaged scripts.

Fine-tune a TAO model with standard supervised training, evaluation, and export, with AutoML bypass and platform-specific credential intake.

Write high-performance GPU kernels using cuTile's tile-based programming model with validation and optimization, including deep agent orchestration for complex multi-kernel tasks.

Scans a DICOM series folder to extract header metadata and produce a preflight verdict without decoding pixel data.

Install Holoscan SDK v4.3+ via Conda in a CUDA 13 environment, including Python bindings and C++ development headers.

Provides storage and environment conventions for NeMo-RL agents on Brev instances, ensuring large experiment outputs go to /ephemeral and secrets are loaded from .env.

Fine-tune NV-Segment-CT VISTA3D on CT NIfTI labels for smoke testing or dataset adaptation, wrapping the upstream MONAI bundle entrypoint.

Trains, evaluates, exports, and runs inference for NVPanoptix3D models that perform panoptic 3D scene reconstruction from posed RGB images, producing 3D panoptic segmentation with occupancy completion.

Train, evaluate, export, prune, and run inference for PointPillars 3D object detection models from LiDAR point clouds using NVIDIA TAO.