Skill UI

SGLang is a high-performance serving framework for LLMs and VLMs that uses RadixAttention prefix caching to deliver structured outputs (JSON/regex/grammar) and agentic workflows with function calls, providing up to 5× faster inference than vLLM across multi-GPU production deployments.

This pattern provides established best practices for implementing efficient similarity search in production systems. It is crucial for building semantic search engines, implementing Retrieval-Augmented Generation (RAG), creating advanced recommendation systems, and optimizing retrieval performance when scaling to millions of vectors.

SimPO (Simple Preference Optimization) is a state-of-the-art, reference-free method designed for aligning Large Language Models (LLMs) using human preference data. It is an efficient alternative to DPO and PPO, notably outperforming DPO without requiring a separate reference model. It is ideal for practitioners who need faster, simpler, and more resource-efficient fine-tuning for preference alignment.

TensorBoard delivers a unified interface to track training metrics, debug network behavior, compare experiments, inspect model graphs, and profile performance for PyTorch or TensorFlow workflows.

Enables Claude to manage AI/ML model versions through the model-versioning-tracker plugin, automating logging, lineage tracking, performance summaries, and version retrieval whenever users request registry or deployment support.

TransformerLens is the standard library for performing mechanistic interpretability research on large language models. It provides clean interfaces to inspect and manipulate transformer internals—such as attention patterns, residual streams, and MLP outputs—via activation caching and HookPoints. This tool is essential for reverse-engineering model algorithms, performing causal tracing, and studying the internal circuits responsible for model behavior.

Guides Claude to automatically tune machine learning model hyperparameters via grid, random, or Bayesian search, generating runnable Python code, executing validations, and reporting best configurations plus performance metrics for accuracy or other goals.

Guide to optimizing TwinMind transcription, balancing accuracy and speed through audio preprocessing, scenario-specific model configs, streaming tweaks, and caching to cut redundant work.

Provides triage, mitigation, communication, and postmortem guidance for resolving Vast.ai integration outages and keeping stakeholders informed.

Specializes in provisioning Vertex AI infrastructure with Terraform, covering Model Garden, Gemini endpoints, vector search indexes, ML pipelines, and enterprise-grade production deployments.

Gathers and analyzes competitor YouTube channels, trending topics, and content gaps to inform a creator’s strategy with data-backed opportunities, channel metrics, and structured reporting.

Reasoning-Driven Image Briefs skill transforms agent prompts into logic-based Gemini 3 style instructions for muapi.ai, adding structured creative briefs (subject/action/context/composition/lighting), precise text cues, and physics-grounded constraints before invoking generation scripts.