Axilla — an AI framework for TypeScript
Axilla is an end-to-end framework to develop LLM applications for enterprise
tl;dr Axilla is a framework for the AI engineer of the future: a product-focused TypeScript developer. Axilla provides a family of modular libraries, which can be incrementally adopted, and together form an end-to-end opinionated framework for LLM development.
The team
Hey there 👋, we’re @Ben Reinhart and @Nicholas Charriere.
We first started working together on ML platforms at Cruise, a leading self-driving car company, where we spent 4 years on the ML platform team. This is also where we became great friends!
At Cruise, we’ve seen the productivity losses that result from a fragmented ML organization. We then spent years designing and building a standardized framework and personally experienced the productivity superpowers that resulted from the framework’s adoption.
🤨 The problem
An accelerating trend in the AI landscape is the shift from in-house model training to foundational models abstracted away behind APIs (chatGPT, Midjourney, llama2…).
A consequence of this trend is that AI teams are morphing from Python-focused data scientists to resemble typical product engineering teams: full-stack web developers increasingly coding AI applications in TypeScript.
We’ve spoken to a number of enterprise customers who’ve confirmed their product teams are now responsible for designing and launching new AI initiatives, especially ones involving LLMs. However, given how recent the explosion of LLMs is, these product engineers don’t yet have the tools to ship AI features quickly and with confidence.
💡 The solution
Axilla provides a family of modular and coherent libraries which together form an end-to-end opinionated framework for LLM development. It is aimed at TypeScript developers who want to build and operate production-grade AI applications.
It’s early, but in the last three weeks we launched our first two modules:
- axgen is focused on data ingestion and retrieval
- axeval is focused on continuous evaluation & quality monitoring
Next, we are excited to begin work on model serving and monitoring, which are two of our most critical components. Together, these integrate production data back into the continuous evaluation pipelines, enable enterprises to run open-source models internally, and allow developers to stop flying blind when it comes to their AI applications.
Demo
As part of our first release, we built a proof of concept that demonstrates how to build a “chat with your documents” application using axgen.
Asks
Try axilla today! It’s open source and available on npm. If you like it, give it a star ⭐️, and come join the conversation on twitter.
Are you a product engineer building AI applications (especially with TypeScript)? We’d love to chat!