Inspiration
It is hectic trying to solve bugs at 3:00am when you have no clue on what broke. We end up wasting a lot of time trying to pinpoint the exact line of code that broke production software or even personal projects. Hence debugging becomes a nightmare
What it does
Aegis fetches real time data from Dynatrace, and uses the provided information to query it's parent brain gemini to hypothesize on possible solutions to the problem. It also tracks commit history to be able to find out exactly which commit broke the system, and suggest a temporary rollback on the previous commits so that the error is fixed while the system keeps running.
How we built it
We used Next.js for the frontend to ensure a user friendly UI, and python to make API calls to the Gemini LLM. We used dynatrace to study our error ridden software "Terra" so that the agent could perform on a real issue and solve an issue in real time. The workflow consists of 5 nodes that ensure the agent does not hallucinate during its tasks. It also has a built in human-in-the-loop barrier to ensure it does not proceed with unsupervised actions.
Challenges we ran into
Hosting on cloud run. Hence the model works locally.
Accomplishments that we're proud of
Being able to design our first AI agent was very tricky, difficult, yet at the same time it was beautiful and educative.
What we learned
How to make an AI agent. How efficiently call LLMs and how to make a full scale system. We also learnt a lot about google architecture which is very friendly to the developer, more so those new in AI engineering
What's next for Aegis
To be able to make a live system that can be used by developers internationally, and have act as a barrier against unprecedented bugs. We aim for Aegis to become a major stakeholder in the future of software development
Log in or sign up for Devpost to join the conversation.