Sentinel AI

Sentinel AI aims to assist developers by identifying the root cause of the CI/CD pipeline failure, generating the patch for the problem, committing (human approval), and creating a new merge request.

Comment

Inspiration

The struggle with software deployment is not something new. Developers always try to push their code to their project repository, and there is a very high chance that the CI-CD pipeline breaks for various reasons.

What it does

Sentinel AI tries to identify the root cause that caused the pipeline failure from the logs, then accordingly creates a "patch" to fix/address the issue. It then waits for the developer/engineer to review the patch and, accordingly, either "Accept" the patch, which would then lead Sentinel AI to create a branch, edit the files, push the changes, and create a merge request, or the developer can "Reject" the patch.

How we built it

We chose to move forward with Google ADK through Python for agent orchestration, FastAPI as the backend technology, and NextJS as the frontend technology, and develop the application through the coding approach. Since we did not have much experience with Google ADK, our instinct for the first couple of days was to read the documentation and go through tutorial videos.

During development, we did take help from LLMs for logic refining, guidance, extensive debugging, and frontend development. However, about 70-75% of the development was our own effort, through documentation, forums, and tutorials.

Challenges we ran into

There were two big challenges for us that we faced during the development of the project: 1) Getting comfortable with Google ADK, as we had relatively little (almost no) experience with Google ADK. 2) The architecture and logic development, of course, was indeed a challenge for us. Besides this, the extensive debugging of getting the project to work is always a challenge, regardless of the experience.

CRITICAL ISSUE: Due to ongoing region-specific payment processing restrictions (Error OR_BACR2_44) with regional Indian debit cards on Google Cloud Console billing, the live environment could not be hosted publicly. However, the entire stack functions perfectly on localhost, as demonstrated step-by-step in our exhaustive 3-minute video overview.

Accomplishments that we're proud of

I really think that coming up with an actual working product built with a new technology (for us) is a big accomplishment for us. Despite all the complexities, we are proud that we understood and tackled the problems, and hence, came up with a working industry-grade project rather than just vibe-coding the entire thing. I personally consider this a big accomplishment and achievement for ourselves.

What we learned

We learned a lot of things from this project. 1) We learnt a new and amazing technology, which is Google ADK. 2) We learnt how an industry-grade project is built. 3) We learnt how to approach a problem, and hence, learnt more about debugging skills.

What's next for Sentinel AI

As of now, I think of this project to be developed as an AI-Based SRE Engine. Besides this, there is no major idea that has yet struck our minds regarding further improvement of this project. But of course, we would definitely pursue this project further if there is a major scope that we can find or define.

Built With

  • fastapi
  • fastmcp
  • google-adk
  • nextjs
  • python
Share this project:

Updates