yif wong

Python, TypeScript, Go, PostgreSQL, prompt engineering, LLM API integration, prompt caching, cost optimization.

Prompt caching and cache-aware routing across Qwen, GLM, Kimi and DeepSeek — how to cut inference cost by routing against cache hit rates.

Adzuki AI — an OpenAI-compatible gateway for Chinese frontier models. Right now: cache-aware routing to cut inference cost.

Happy to talk model pricing, prompt-cache economics, and building on LLM APIs. If you're weighing which Chinese frontier model fits your workload, that's a good reason to say hey.