LLM integration

Description

Build, advise on, and review application code that calls LLM APIs (any provider — Anthropic, OpenAI, Google Gemini, open models) — server-side key handling, model selection via config, structured output with JSON Schema and runtime validation, tool/function calling, streaming to the UI, timeouts/retries/backoff on rate limits, token and cost budgeting, caching, prompt-injection defenses, PII handling, evaluation, and cost-safe testing. Use whenever the code imports an LLM SDK (@google/genai, @anthropic-ai/sdk, openai, ai) or the user adds AI features like summarization, classification, extraction, chat, or agents to an app — and before running any loop of live model calls.