I recently rebuilt the "Virtual Me" chatbot on this site, which answers questions about my background as if it were me. The old version ran on Flask, OpenAI, and a retrieval-augmented-generation ("RAG") pipeline that made sense in 2023, when context windows were much smaller. Jumping to the advances made since then, the architecture was clearly overkill for a corpus of this size. The new system is stripped down to just FastAPI and Claude. My resume and work history load straight into the model's context on every request; retrieval from a vector database is no longer necessary. It's now a much smaller system, cheaper to run.
This blog is part of the same rebuild. Posts here are short, written the same file-based way I update my resume. The recent ones feed straight into the chatbot's own knowledge base, so if you ask it what I've been writing about lately, it'll actually know.
