Hello, world

I'm starting this blog to write about the practical side of building AI systems — the parts that don't make it into a demo or a paper: fine-tuning speech models for languages with messy data, debugging why an agent works in eval but not in production, and the infrastructure decisions that end up mattering more than the model choice.

Most of what I'll write about comes directly from what I'm working on: voice AI (STT/TTS/S2S), GenAI and LLM fine-tuning, RAG pipelines, and the systems and product decisions around shipping AI features that real users depend on.

More soon.