Skill UI

Provides production-ready Megatron-Core workflows for training 2B–462B parameter LLMs with tensor/pipeline/expert/context parallelism, maximizing GPU efficiency on H100 clusters.

Guides training large language models with Volcano Engine's verl library, covering PPO/GRPO workflows, distributed backends, multi-turn rollouts, reward configs, critic-enabled PPO, and production deployments using Ray workers.

This skill guides the development of advanced voice agents, representing the frontier of natural human-AI interaction. It covers both low-latency Speech-to-Speech (S2S) architectures (like OpenAI Realtime API) and fully controllable STT→LLM→TTS pipelines. Key focus areas include minimizing end-to-end latency (targeting <800ms), handling conversational nuances, background noise, and managing interruptions for applications like customer support, IVR, and live translation.

Guides building async worker pipelines for real-time conversational voice agents with streaming transcription, LLM reasoning, and interrupt-aware TTS across providers.

A comprehensive guide for developing, optimizing, and standardizing skills for AI agents. This resource provides a decision tree for choosing the right architecture (Tier 1-3), incorporating best practices like CSO (LLM SEO) and Anti-Rationalization, and ensuring compliance through rigorous testing and checklists. Essential for robust AI system building.

Hypogenic automates LLM-driven hypothesis generation and testing over tabular datasets, blending literature insights with empirical analysis to accelerate discovery in domains like deception detection or AI content evaluation.

MarkItDown converts PDFs, DOCX, PPTX, XLSX, HTML, JSON, XML, archives, YouTube links, images via OCR, and audio via transcription into clean Markdown, providing token-efficient text ready for LLMs and documentation workflows.

AI-powered search using Perplexity models via LiteLLM and a single OpenRouter API key for real-time, web-grounded answers with citations, ideal for up-to-date research, literature surveys, fact checking, and domain-specific queries beyond model cutoffs.

A specialized tool for generating long-form content, including guides, blogs, and tutorials, that maintains a unique, human-like brand voice. It ensures structural integrity and factual accuracy, preventing the generic, boilerplate feel often associated with raw LLM output. Ideal for content teams needing sophisticated, reliable narrative development.

A powerful tool to scaffold and ship production-ready Model Context Protocol (MCP) servers directly from OpenAPI specifications. It supports Python and TypeScript, ensuring schema validation, safe evolution, and robust tool definition generation, eliminating the need for manual wrapper coding. Use it to expose existing REST APIs to LLM agents or build shared tool integrations.

A comprehensive toolkit designed to professionalize LLM prompt workflows. It moves prompts from ad-hoc drafts to measurable, production-ready assets through A/B testing against structured test cases, immutable version tracking, and governance guidelines. Essential for marketing teams needing verifiable quality and safety for AI-generated content.

A comprehensive guide covering patterns and architectures for automating code execution using LLMs. It spans simple sequential pipelines (e.g., daily dev scripts) to advanced, RFC-driven multi-agent Directed Acyclic Graph (DAG) orchestration. Ideal for building CI/CD style continuous development pipelines, managing complex state, and automating quality gates without human intervention.