I was an early enthusiast of both LangChain and LlamaIndex (and I wrote a book using both frameworks, free to read online [1]) but I had some second thoughts wh

I was an early enthusiast of both LangChain and LlamaIndex (and I wrote a book using both frameworks, free to read online [1]) but I had some second thoughts when I started when I started writing LLM examples for my Common Lisp and Racket books that were framework-free, even writing simple vector data stores from scratch. This was, frankly, more fun. For my personal LLM hacking in Python, I am starting down the same path: writing simple vector data stores in NumPy, write my own prompting tools and LLM wrappers, etc. I still think that for many developers LangChain and LlamaIndex are very useful (and I try to keep my book up to date), but I usually write about things of most interest to me and I have been thinking of rewriting a new book on framework-free LLM development. [1] https://leanpub.com/langchain/read https://leanpub.com/langchain/read