Inspiration
The codesource presentation today got me thinking about software development principles, and the challenges in AI code gen for large projects and iterative development.
Software development principles reduce the attention costs for humans, so why not for LLMs too?
What it does
This tool allows for editing codebases with AI. Users ask for a change they want made to their software, and the Agents coordinate to make it.
How we built it
I wrote this using Camel as my agent framework, with a bunch of help from Claude.
Challenges we ran into
It was challenging to figure out how agents should split responsibilities and communicate.
Concurrency has been a struggle too. It currently gets stuck after the first request.
The version of my codebase at the time of submission can be found here https://github.com/CaelumF/encapt/commit/d3e9e3b0c4f0bb4b347d75b5da809bea4116171a If you are curious though, you can look at the later versions, since I will keep updating it.
Accomplishments that we're proud of
Safety - Restricting agents to writing files in a single directory reduces data loss risk
What we learned
I learned about Camel's upcoming workforce potential.
What's next for Encapt
Make it work fully. Also I want to integrate Galileo.
Built With
- camel
- python
Log in or sign up for Devpost to join the conversation.