Data analysis, machine learning, LLM apps, and prompt design (9.1K skills)
Build and troubleshoot a GPU-powered service that checks factory videos for missing or out-of-order procedure steps. Guides video segmentation, action classification with a vision-language model, deployment, evaluation, and latency measurement for files and live cameras.
Write and debug nvMolKit Python code for processing large molecule batches on GPUs. Covers molecular fingerprints, compact representations of molecular features, plus similarity calculations, 3D conformer generation, optimization, clustering, and molecular searches using RDKit objects.
Pretrain a new KERMT chemistry model from a molecular corpus, starting with random weights. The skill validates the data, builds fresh vocabularies, estimates runtime, and launches containerized training with logs and records for reproducing the run.
Generate molecular predictions from a CSV of SMILES strings, a text notation for molecular structures, using a finetuned KERMT model. Validate inputs, clean data, compute molecular features, and return predictions with execution logs and a replay command.
Generate DNA sequences with NVIDIA Evo 2, check their length and alphabet, and save sequence files alongside probabilities and timing metrics. A local deployment also supports extracting numerical outputs from selected internal model layers.
Extend an existing grover_base model with a new decoder that reconstructs molecular structure strings, then continue pretraining on your corpus. The workflow validates inputs, rebuilds vocabulary components, and records the conversion and training outputs.
Fine-tune a pretrained KERMT model on a CSV of molecular structures and labeled targets. The workflow checks inputs, prepares features, launches training in a container, and saves trained models, logs, and test results.
Move pandas-style table processing to NVIDIA GPUs using cuDF. Choose minimal-change acceleration or explicit GPU code, handle joins and aggregations, compare results with pandas, and use dask-cuDF when data exceeds GPU memory.
Build English clinical speech-recognition benchmarks by selecting medical terms, checking pronunciations, and generating synthetic audio. Outputs include audio files and a NeMo manifest, a structured file listing recordings and metadata, ready for a separate scoring stage.
Predict biomolecular structures and estimate how strongly a ligand binds using Boltz2 NIM. Submit proteins, DNA, RNA, and ligand inputs through NVIDIA's hosted API or local Docker, then save structures and inspect confidence and affinity scores.
Design protein backbones, the structural framework of proteins, with RFDiffusion through NVIDIA NIM. Generate new structures, build around existing structural motifs, or design binders, then save PDB files for subsequent amino acid sequence design.
Fine-tune Parakeet TDT v2 on clinical vocabulary using NeMo, then compare key-term error rates before and after training. The results guide whether to keep the model, expand the dataset, or prepare it for deployment.
Predict protein structures by connecting MSA-Search with OpenFold3. It finds related sequences, aligns them to provide evolutionary context, and saves predicted structures with confidence scores. It also supports comparing predictions with and without that context.
Turn molecules listed in a SMILES CSV into numerical feature vectors using a KERMT model. The workflow checks the model and input, cleans molecular notation, and saves four sets of vectors with validity information.
Predict structures of proteins, DNA, RNA, and complexes with small molecules using OpenFold3. Guides hosted API requests or local Docker setup, saves structure files, and helps interpret confidence scores supplied with the predictions.
Write, review, and migrate DALI data processing code using dynamic mode, which calls operators directly from ordinary Python. Handle readers, batches, GPU processing, PyTorch conversion, random state, and checkpoints with the conventions specific to this API.
Extract selected metadata from a single DICOM medical imaging file into JSON, including scan type and study details. Flag personal health information in a limited set of standard tags for technical inspection.
Predict how small molecules may bind to proteins using DiffDock through NVIDIA NIM. Prepare molecular inputs, run hosted API or local Docker requests, and save ranked binding poses with confidence scores for inspection.
Build DeepStream commands for image and video inference, object detection, tracking, and streaming. The skill asks about missing requirements, assembles a GStreamer processing pipeline, and checks its syntax, components, properties, and connections where supported.
Run DeepStream's Multi-View 3D Tracking reference application on bundled samples or synchronized MP4 recordings. The skill guides setup, camera calibration integration, annotated and bird's-eye visualizations, saved video outputs, Kafka tracking metadata checks, and run cleanup.
Build and solve routing problems in Python with cuOpt, including delivery fleets, round trips, and pickup-delivery tasks. Define vehicle capacities, arrival windows, and visit order constraints, then inspect routes, total cost, and orders that cannot be served.
Create a Python configuration for generating synthetic datasets with the Data Designer library. Describe the data you need, then refine requirements interactively or let the agent choose assumptions and assemble the generation pipeline.
Integrate a supported object detection model from HuggingFace or NVIDIA NGC into NVIDIA DeepStream. Automates model acquisition and conversion, TensorRT engine building, detection output parsing, multi-stream performance benchmarks, and reports with charts in Markdown, HTML, and PDF.
Convert a single CT DICOM series into a NIfTI volume with Hounsfield units, the CT intensity scale. A JSON summary records image dimensions, spacing, orientation, and the coordinate transformation matrix for checking the conversion.