Bryan Banner

I build the stuff underneath AI applications. Orchestration engines that route across multiple LLMs and know when to fall back. Embedding pipelines that actually understand code structure instead of blindly chunking by line count. RAG systems that detect their own drift and reindex before they start hallucinating.

I spent the first half of my career in network engineering and systems programming. That means when something breaks in an AI pipeline, my first instinct is not to tweak the temperature parameter. I want to know what is happening at the syscall layer, whether the vector store is actually returning the right embeddings, and if the routing logic is making sane decisions or just burning API tokens.