Skip to main content

LangChain & OpenAI

Today I learned how the integrate LangChain into the OpenAI API. Since it was quite a bit to wrap my head around, I also had to do the deep dive on some of the fundamentals of machine learning and vector databases. I published a simple demo and started making notes about the process.

I've started making notes and related resources @ /docs/sass/openai 1. I also put up a simple demo @ https://jenks.davidawindham.com 2 that uses a doctoral dissertation from a friend who teaches computer science locally. It's powered by the OpenAI API, LangChain, and a Pinecone vector database.

I'm about halfway done with a course ChatGPT Prompt Engineering for Developers3, and I spent a couple hours on the phone talking with a fellow who has a very deep understanding of machine learning and artificial intelligence. It was mostly me asking him to explain some of these to me:

  • why did my vector database use a cosine metric and what does trigonometry have to do with machine learning?
  • what are vector dimensions and how to they related to machine learning
  • how are artificial neural networks built
  • how are linear regressions used in machine learning
  • how are weights and dimensions added to models.

It's all quite a bit to take in. My brain was lit 🔥 on LangChain, so I spent a good hour just decompressing. I still feel a bit like I did when I first started learning programming in that I have such a naive knowledge of the fundamentals. In the last couple of weeks, I've spent a good bit of time just getting reacquainted with Python development environments like Jupyter, Colab, Tensorflow, and PyTorch. I don't think I'll ever be building neural networks, but I would at least like to know how it's done so that the training, chaining, prompting, transformers, and pipelines will become easier for me to use.

Although I can see some really practical applications, I think my first custom project will likely be an absurd AI chatbot built on top of my personal chats, emails, calendar, this knowledge-base, and my website4. My wife and I were discussing AI a couple days ago and our biggest takeaway is that when systematic knowledge like that of a Jeopardy contestant becomes less impressive, the more that our personal human qualities like emotion and artistic creativity become important.

  1. Docs / SAAS / OpenAI - /docs/saas/openai
  2. TedBot - https://jenks.davidawindham.com
  3. ChatGPT Prompt Engineering for Developers - https://learn.deeplearning.ai/chatgpt-prompt-eng/
  4. A Second Brain - https://davidawindham.com/a-second-brain/