The really interesting part: your server isn't the main problem

Your initial server response is around:

454 ms

and TTFB in the LCP breakdown is:

450 ms.

That's not fantastic, but it's nowhere near the reason you're seeing:

FCP: 22.8s

LCP: 27.1s

The report identifies 3.09 seconds of element render delay in the LCP breakdown, while the overall load is much worse because of the broader critical rendering/resource chain.

In other words:

Don't start by moving AIToolsHaven away from Vercel or rebuilding your backend.

Your biggest gains are on the frontend delivery path.