Demystifying LCEL & LangChain

I love the holiday season. There's family around, sales and deal making is in high gear, and things generally slow down enough for me really dive into a new piece of technology. This year is of course the year of AI and hacking on AI. There's a ton of content out there that covers what cool AI startups are around, or how you can construct the perfect prompt, but what I'm curious about is diving into building applications with AI. What's that experience, how are the communities, and how can I get a production app live?

Focused has already spent a good amount of time and energy investing in AI, if you haven't seen it check out our AI Knowledge Base on GitHub it's a powerful template to get you up and running with a custom RAG in no time. You can play with our version on https://chat.withfocus.com/. But, I don't want to talk about all that I want to get technical so let's dive into some LangChain LCEL.

First: What is LCEL

LangChain Expression Language, LCEL is a new compossible interface for building AI applications with LangChain. It offers a few really convenient features right out of the box, and I would recommend using it for any new application development. It's even likely that LangChain themselves will deprecate any other way of building chains.

What do you get with LCEL:

Streaming support
Async support
Parallel execution
Retries and fallbacks
Access intermediate results
Input and output schemas
And a bunch of LangSmith/LangServe Support

That's pretty great considering streaming and async execution are some of the most sought after features for any AI application. People just want to make it feel like ChatGPT.

Cool, that's what LCEL is, but really what is it? To me, LCEL is a clean and clear interface to having each part of your AI application work together. It's a pipeline for prompt creation, LLM interaction, and output formatting. It's simple, it's clean, and it's insanely powerful.