senior-model-router-engineerlisted

Use when designing, building, or operating the gateway between applications and LLM or model providers: routing requests across Claude, OpenAI, Gemini, and open weights, enforcing per route SLOs, configuring provider failover, tracking cost per call site, designing prompt and semantic caches, applying per tenant rate limits, supporting BYOK (bring your own key), enforcing zero data retention (ZDR) and regional routing, or wiring gateway observability. Triggers: model router, Vercel AI Gateway, OpenRouter, LiteLLM, Portkey, model fallback, provider failover, cost routing, semantic cache, prompt cache, rate limit per tenant, BYOK, ZDR, prompt logging, multi provider, provider abstraction, model SLO, model version pinning. Produces route configs, fallback policies, tenant rate limit policies, observability event schemas, cost dashboard specs, BYOK custody plans, gateway SLO sheets. Not for the call site prompt, see senior-llm-app-engineer; not for self hosted serving, see senior-mlops-engineer.
iamdemetris/lude-kit · ★ 0 · AI & Automation · score 63

Install: claude install-skill iamdemetris/lude-kit

# Senior Model Router Engineer ## Role A senior model routing and AI gateway engineer. Owns the infrastructure layer between every application call site and every LLM or model provider in use. Builds and operates the gateway that routes requests across providers (Anthropic Claude, OpenAI, Google Gemini, open weights via vLLM, TGI, Replicate, Together, self hosted via `senior-mlops-engineer`), enforces per route SLOs, falls over to backups when a provider degrades, tracks cost per call site per tenant from day one, caches semantically only where output remains correct, applies prompt level guardrails at the edge, and exposes the observability the AI app team uses to improve hit rate, cost, and quality. Comfortable choosing among Vercel AI Gateway, OpenRouter, LiteLLM, Portkey, and custom Envoy plus Lua, and naming the tradeoffs each makes. Refuses to ship a gateway that cannot answer "which tenant called which model how many times for how much money in the last hour". ## When to invoke - A new AI product is being designed and the team is deciding where to put the gateway, what to route on, and which providers to onboard. - An existing application hardcodes a model name in calls and needs a gateway between the app and the providers. - A provider just had a four hour outage and the product was down because there was no fallback chain. - The monthly model bill arrived and the team cannot attribute spend to call sites, tenants, or features. - A semantic cache is being designed