Show HN: AI-patterns – 20 TypeScript patterns for production AI apps (github.com)

ai-patterns is a new npm library that bundles 20 battle-tested TypeScript patterns for building production-ready AI applications—retry, circuit breaker, timeout, rate limiter, fallback (cross-provider failover), cache/memoize, fan-out/concurrency, human-in-the-loop, smart context-window management, A/B testing, cost tracking, prompt versioning, response validation, idempotency, saga, and more. It targets the common pain points of AI integrations (fragile retry logic, lack of circuit breakers, uncontrolled costs, rate limits, and boilerplate) by giving developers composable, type-safe building blocks with zero runtime dependencies. Technically, ai-patterns is built for strict TypeScript (full generics and inference), offers observable lifecycle callbacks for monitoring, and provides stateful primitives (defineCircuitBreaker, defineRateLimiter) that you can reuse across calls. Patterns are composable via compose/withRetry/withTimeout middleware, support custom backoff strategies (exponential, linear, fixed), AbortSignal-based timeouts, concurrency control in fanOut, and cross-provider fallback (OpenAI → Claude → Gemini). Examples and runnable snippets demonstrate practical combos (e.g., response validation + retry + timeout, A/B testing with cost tracking). Packaged MIT-open and inspired by Vercel AI SDK’s DX, it’s designed to reduce boilerplate and harden AI workflows for production use.