Skill UI

A comprehensive guide to optimizing deep learning workflows and maximizing efficiency on cloud GPU platforms like Vast.ai. Learn techniques for selecting optimal instances by performance-per-dollar, reducing startup latency, tuning data pipelines (DataLoader, pin_memory), and scaling multi-GPU training using PyTorch and Bash scripting.

This comprehensive guide outlines best practices for optimizing vector indexes for production systems. It covers critical techniques such as tuning HNSW parameters, implementing quantization strategies, and optimizing memory usage. Use this when you need to balance search latency, recall, and scalability for large-scale vector databases.

A comprehensive guide to building and deploying Vercel Edge Functions. These functions run on Vercel's global Edge Network (V8 isolates), providing ultra-low latency and zero cold starts. Learn how to use Web Standard APIs for critical tasks such as authentication, geolocation routing, A/B testing, and high-performance API middleware, drastically improving application speed and user experience.

This guide provides comprehensive methods for load testing and capacity planning of Vercel serverless deployments. Learn how to use professional tools like k6 and autocannon to simulate realistic traffic spikes, measure latency (P95, P99), analyze cold start impact, and tune concurrency settings (Fluid Compute) to ensure optimal application performance and scalability under heavy load.

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.

Comprehensive expertise in designing and implementing low-latency, production-ready voice AI applications and conversational agents. Covers the full stack from real-time audio streaming (WebRTC, LiveKit) to advanced NLU and speech synthesis (OpenAI, ElevenLabs, Deepgram). Ideal for building next-generation voice assistants, virtual call centers, and voice-enabled mobile experiences while optimizing for perceived responsiveness and latency budgets.

Expert resource for designing, deploying, and optimizing serverless applications on the Cloudflare Edge Network. Covers advanced topics including Workers, KV, D1, Durable Objects, and R2 storage. Ideal for building high-performance, low-latency full-stack applications and implementing edge security features.

This skill monitors the performance of Claude Code and the local operating system. It diagnoses slowdowns, measures CPU/RAM/Disk usage, checks API latency, and generates detailed system health reports. It helps determine if the bottleneck is local hardware or the remote API connection, providing actionable corrective suggestions to improve the development workflow.

A framework for building production-ready, low-latency, multimodal AI agents. It allows developers to orchestrate complex pipelines that seamlessly handle real-time audio, video, and text data. Key capabilities include integrating multiple AI services (LLMs, STT, TTS), managing conversation context, and enabling tool use and function calling for advanced, interactive AI applications.

This blueprint guides the creation of F.R.I.D.A.Y., a tactical, low-latency voice assistant inspired by Iron Man. It utilizes the specialized Pipecat framework to orchestrate a comprehensive, real-time audio pipeline: STT (OpenAI Whisper/GPT-4o) -> LLM (Gemini) -> TTS (OpenAI Nova). Ideal for developers needing to build advanced, multi-provider, conversational voice applications.

This skill guides the implementation of envelope encryption using AWS Key Management Service (KMS). Envelope encryption is a robust security pattern where data is encrypted locally using a Data Encryption Key (DEK), and the DEK itself is protected by a master Key Encryption Key (KEK) managed by AWS KMS. This approach overcomes the limitations of direct KMS encryption, enabling unlimited data size encryption, reducing latency, and allowing offline encryption capabilities by keeping the master key secure in an HSM.

Provides end-to-end guidance for fine-tuning and serving Physical Intelligence’s OpenPI pi0 family (pi0, pi0-fast, pi0.5) with both JAX and PyTorch backends, covering dataset prep, norm stats, checkpoint conversion, and low-latency WebSocket policy inference for ALOHA, DROID, and LIBERO tasks.