Skill UI

Bridges frozen vision encoders and LLMs for high-quality image captioning, VQA, retrieval, and multimodal chat, enabling zero-shot reasoning without fine-tuning and offering efficient Q-Former training.

This skill set provides advanced, structured strategies for optimizing Large Language Model (LLM) interactions. It covers techniques such as intelligent context summarization, dynamic trimming, context routing, and precise token budget allocation to prevent 'context rot'. It is essential for building robust, long-running, multi-turn conversational AI systems.

Defines production-ready Deepgram SDK usage for TypeScript and Python, covering pre-recorded transcription, live streaming, diarization, multi-language handling, and error responses to guide integrations and team standards.

A comprehensive toolkit for evaluating the performance of code generation models. It supports benchmarking against industry standards like HumanEval, MBPP, and MultiPL-E across multiple languages. Use this tool to quantitatively compare different LLM coding abilities and measure code synthesis quality using pass@k metrics.

LLaVA is an open-source vision-language assistant pairing CLIP vision encoders with Vicuna/LLaMA models to power multimodal chatbots, multi-turn image Q&A, captioning, and instruction-following tasks.

Guides designers of language-model multi-agent systems through supervisor, swarm, and hierarchical patterns while stressing context isolation, coordination protocols, and token economics for complex research or tool-intensive workflows.

SentencePiece provides a lightweight, deterministic subword tokenizer that trains on raw Unicode text and scales across multilingual and CJK applications without language-specific preprocessing.

Guides RLHF-style LLM fine-tuning with verl's HybridFlow stack, scaling PPO/GRPO/DAPO across FSDP/Megatron/vLLM backends for math reasoning, agentic multiturn, and vision-language workflows.

Whisper Speech Recognition by OpenAI provides multilingual speech-to-text, translation, and language detection, handling noisy audio and podcasts with model sizes for every speed/quality tradeoff. Use for transcription, translation, meeting notes, and batch workflows.

Build, run, and visualize multi-step AI generation workflows via muapi.ai. Translate natural language intents into runnable node graphs, discover or edit existing pipelines, and execute them interactively or through the CLI with automated monitoring.