AI provider abstraction for a flexible RAG system with Ollama and OpenAI
I could have built Ollama directly into the backend. I could have built OpenAI directly into it. Instead, I built an abstraction layer that allows both, and understood why this is not overengineering but the only strategically sound decision.