Loading...
Loading...
Found 32 Skills
Create an AI Evals Pack (eval PRD, test set, rubric, judge plan, results + iteration loop). Use for LLM evaluation, benchmarks, rubrics, error analysis/open coding, and ship/no-ship quality gates for AI features.
Use when evaluating LLMs, running benchmarks like MMLU/HumanEval/GSM8K, setting up evaluation pipelines, or asking about "NeMo Evaluator", "LLM benchmarking", "model evaluation", "MMLU", "HumanEval", "GSM8K", "benchmark harnesses"
Build a structured taxonomy of failure modes from open-coded trace annotations. Use this skill whenever the user has freeform annotations from reviewing LLM traces and wants to cluster them into a coherent, non-overlapping set of binary failure categories (axial coding). Also use when the user mentions "failure modes", "error taxonomy", "axial coding", "cluster annotations", "categorize errors", "failure analysis", or wants to go from raw observation notes to structured evaluation criteria. This skill covers the full pipeline: grouping open codes, defining failure modes, re-labeling traces, and quantifying error rates.
Evaluate LLM systems using automated metrics, LLM-as-judge, and benchmarks. Use when testing prompt quality, validating RAG pipelines, measuring safety (hallucinations, bias), or comparing models for production deployment.
Design LLM-as-Judge evaluators for subjective criteria that code-based checks cannot handle. Use when a failure mode requires interpretation (tone, faithfulness, relevance, completeness). Do NOT use when the failure mode can be checked with code (regex, schema validation, execution tests). Do NOT use when you need to validate or calibrate the judge — use validate-evaluator instead.
Calibrate an LLM judge against human labels using data splits, TPR/TNR, and bias correction. Use after writing a judge prompt (write-judge-prompt) when you need to verify alignment before trusting its outputs. Do NOT use for code-based evaluators (those are deterministic; test with standard unit tests).
Create diverse synthetic test inputs for LLM pipeline evaluation using dimension-based tuple generation. Use when bootstrapping an eval dataset, when real user data is sparse, or when stress-testing specific failure hypotheses. Do NOT use when you already have 100+ representative real traces (use stratified sampling instead), or when the task is collecting production logs.
Audit an LLM eval pipeline and surface problems: missing error analysis, unvalidated judges, vanity metrics, etc. Use when inheriting an eval system, when unsure whether evals are trustworthy, or as a starting point when no eval infrastructure exists. Do NOT use when the goal is to build a new evaluator from scratch (use error-analysis, write-judge-prompt, or validate-evaluator instead).
Help the user systematically identify and categorize failure modes in an LLM pipeline by reading traces. Use when starting a new eval project, after significant pipeline changes (new features, model switches, prompt rewrites), when production metrics drop, or after incidents.
Run the Codex Readiness unit test report. Use when you need deterministic checks plus in-session LLM evals for AGENTS.md/PLANS.md.
Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.
LLM observability platform for tracing, evaluation, prompt management, and cost tracking. Use when setting up Langfuse, monitoring LLM costs, tracking token usage, or implementing prompt versioning.