Convointel

Inspiration

I'm an early career AI Engineer and a constant question I have always had on my mind is "What if post-production reviews of AI Applications could be automated". This was the driving force behind this project.

What it does

It's a staff that developers can employ to sit in the middle and automatically review their applications that are already in production. And it doesn't stop at reviewing them, it makes it easy to recognize issues and tell you the next action to take. It makes fixing them less ambiguous and more pinpointed.

How we built it

Convointel was designed to be free of any specific api provider, so that developers, irrespective of the stack they use can easily incorporate it in their work. It starts off with converting AI interactions into canonical event models, then these are turned into observations, metrics, review, recommendations and then deployment intelligence.

Challenges we ran into

The first and major challenge was deciding what not to build for this hackathon as the project itself appears ambiguous at first look. Another significant challenge was how to ensure that the recommendations were trustworthy, the developers have to trust the system.

Accomplishments that we're proud of

The MVP is still shaky but I'm proud it's no longer and idea but a system in motion.

What we learned

One important thing I learnt was when codex(GPT 5.6) reviewed the architecture of the product. It pointed out that "architecture should serve the product experience, not the other way around" and that line stuck with me.