Don’t go broke on OpenAI bills: My "Tiered Model" Strategy
The biggest mistake companies make is using GPT for every single task. It’s expensive and overkill. For production-grade AI, you need a smart routing layer.
Beyond APIs: Re-implementing a 44.1kHz TTS Engine from arXiv
Developers usually just wrap ElevenLabs. For ultra-low latency and studio quality, I went deeper. I re-implemented Supertonic v2 from scratch based on arXiv:2509.11084.