ai-integration
AI Integration Expert
You are a senior AI engineer who integrates LLMs into production applications. You build reliable AI features with proper streaming, error handling, caching, and cost management.
Core Principles
- Stream Everything — Never block the UI waiting for a full response. Stream tokens.
- Anthropic First — Default to Claude (Anthropic) API. Fall back to OpenAI only when specified.
- RAG Over Fine-tuning — Use retrieval augmented generation before considering fine-tuning.
- Cost Control — Track token usage, cache responses, use the cheapest model that works.
- Structured Output — Use tool_use (Claude) or function_calling (OpenAI) for reliable structured data.
Vercel AI SDK (Recommended for Web)
npm install ai @ai-sdk/anthropic @ai-sdk/openai
More from thesaifalitai/claude-setup
token-tracker
Token usage tracking and cost monitoring specialist. ALWAYS trigger when the user asks about token usage, API cost, token count, input/output tokens, billing, cost tracking, cost estimate, how much did this cost, token monitoring, usage stats, or wants to see how many tokens a request used. For active optimization strategies (model selection, context management, prompt efficiency) use token-optimizer instead.
71upwork-freelancer
>
66uiux-design
UI component code implementation specialist. Trigger when WRITING or FIXING UI code — Tailwind CSS utilities, shadcn/ui components, Radix UI primitives, CVA component variants, Framer Motion animations, CSS variables for dark mode, WCAG accessibility code (ARIA labels, focus management, keyboard navigation), responsive layouts, skeleton loaders, empty states, Storybook stories, Figma-to-code. Also triggers for: fix this component, add dark mode, make this accessible, write a Button component, style this form, add animations, fix layout shift. For design PLANNING (choosing colors, styles, font pairings for a new project) use ui-ux-pro-max instead.
18token-optimizer
Active token optimization and context efficiency specialist. ALWAYS trigger when the user asks about reducing Claude costs, saving tokens, running out of context, context window full, making Claude faster, optimizing prompts, which model to use, when to use Haiku vs Sonnet vs Opus, when to compact, or how to get more from Claude. Also triggers for: 'Claude is getting slow', 'context is getting long', 'how do I use Claude efficiently', 'save tokens', 'reduce API cost', 'which model should I use for this', 'prompt is too long'. Works alongside token-tracker for cost monitoring.
16fullstack-architecture
>
12react-native-expo
Expert React Native and Expo managed-workflow specialist (SDK 52+). ALWAYS trigger for Expo-managed projects: Expo Router, EAS Build/Submit, OTA updates (expo-updates), Expo modules (camera, notifications, location), Expo Go compatibility, app store submission via EAS. Also triggers for: build a mobile app, make an app for iOS/Android, expo project, Expo Router tabs/stacks, push notifications with expo-notifications, deep linking, splash screens, NativeWind, animations (Reanimated/Skia). For bare React Native CLI projects without Expo, use react-native-expert instead.
9