Skill UI

FAISS (Facebook AI Similarity Search) is a library designed for efficient similarity search and clustering of dense vectors. It supports indexing and searching on billions of vectors at scale, featuring GPU acceleration and multiple index types (Flat, IVF, HNSW, PQ) to balance speed and accuracy. It is essential for building high-performance, large-scale applications requiring fast k-NN retrieval or pure vector similarity matching, especially in fields like RAG and recommendation systems.

This skill provides capabilities for bidirectional audio processing using fal.ai models. It supports Text-to-Speech (TTS), converting written text into natural-sounding speech, and Speech-to-Text (STT), accurately transcribing spoken audio into readable text. Use this skill for applications requiring high-quality audio input analysis or synthetic voice generation, such as virtual assistants, content digitization, or multilingual communication tools.

This skill enables the generation of high-quality images and videos using advanced fal.ai AI models. It is designed for creative professionals, artists, and developers who require powerful, generative visual content. Use this when your task involves transforming text prompts into sophisticated visual media, making it ideal for concept art, multimedia prototyping, or advanced content creation.

Automatically activates when you mention feature engineering helper, offering guidance on data preparation, model training, hyperparameter tuning, and experiment tracking for ML projects.

Guides ML engineers through dataset prep, PEFT strategy selection, LoRA/QLoRA adapter setup, training monitoring, and deployment checks for practical LLM fine-tuning, quantization, and adapter merge workflows.

Fine-tune LLMs with TRL’s RLHF toolkit, covering SFT, DPO, PPO/GRPO and reward model training so you can align HuggingFace Transformers with human preferences or feedback in post-training workflows.

This workflow connects to the Fireflies.ai API via GraphQL to fetch and process detailed meeting transcripts. It handles speaker-diarized data, extracting structured insights such as action items, key topics, sentiment analysis, and speaker breakdowns. It enables the creation of comprehensive meeting intelligence reports and robust downstream pipelines. Use this when you need to automate the retrieval, summarization, or deep analysis of recorded meeting content.

Automates FlowiseAI workflows by orchestrating Rube MCP and Composio tools; always searches current tool schemas, validates active Flowiseai connections, and executes schema-compliant multi-tool calls with session reuse.

Gastown is a robust multi-agent orchestration system designed for Claude Code. It enables the simultaneous execution of multiple AI agents on complex projects. It manages the entire development lifecycle, from setting up workspaces and tracking work items (beads) to managing agent lifecycles (polecats/crew) and merging final code (Refinery). Use it when coordinating complex, parallel AI tasks.

Generates complete, production-ready Google Cloud Platform (GCP) code examples and infrastructure templates. It spans various modern AI patterns, including ADK agents, Genkit RAG flows, and multimodal analysis using Vertex AI and Gemini. The output includes runnable code, deployment configuration (Terraform/IaC), security best practices (IAM, VPC-SC), and monitoring setup, ensuring enterprise-grade reliability and cost optimization.

A comprehensive guide to building enterprise-grade AI applications using Firebase Genkit. This skill covers the full development lifecycle, including implementing advanced RAG systems, multi-step tool-calling agents, and complex workflows across Node.js, Python, and Go. It teaches best practices for schema validation (Zod/Pydantic), robust error handling, OpenTelemetry tracing, and deploying reliably to Firebase Functions or Cloud Run.

This guide details the use of the GGUF format and quantization techniques for running large language models (LLMs) efficiently. GGUF enables optimal inference on consumer hardware, Apple Silicon, and systems requiring CPU-only operation. By compressing models using various K-quant methods (like Q4_K_M), developers can significantly reduce memory footprint and hardware requirements while maintaining high performance.