Hello! I am a student from Romania and I need your opinion on my web project

Hi everyone! I’m a student from Romania currently developing PLI 7, an AI-driven tool built with Next.js and the Gemini model.

While many AI wrappers rely on rigid, pre-built cloud solutions, I decided to architect my own backend from the ground up. My goal was to move beyond the limitations of standard "out-of-the-box" integrations to ensure maximum stability, lower latency, and more precise control over how the AI responds.

I’m at a stage where I’d love some peer review from the people community. Specifically:

Architecture & Logic: Does a custom backend approach for Gemini-based tools offer the advantages I’m aiming for (stability/speed), or are there hidden pitfalls I should watch out for?

UI/UX Evaluation: How does the interface feel? Is the flow intuitive for a tool of this nature?

Vision: Does this "sovereign backend" approach make sense for a modern AI tool, or is the industry moving elsewhere?

Live Project: https://pli7.vercel.app/

I’m open to any and all technical critiques, suggestions for the stack, or ideas on features. Thanks for your time!

 

Another spammer and teachable moment. lol Turn spam into lemonade.

A “sovereign backend” can absolutely help with stability/speed if you’re aiming for better control over latency, retries, caching, and provider swapping. The hidden pitfalls are mostly operational: rate limits/quotas, streaming reliability, timeouts, backoff/retry storms, and making sure failures degrade cleanly (good errors, partial results, fallback behavior). Observability matters a lot too—logs, tracing, and clear metrics.

UI/UX-wise, tools like this live on clarity and feedback: obvious primary action, visible system status (loading/streaming/errors), and a predictable flow that keeps users from guessing what happens next. If you have multiple modes, keep the first run simple and progressively reveal advanced stuff.

Vision-wise, this approach still makes sense—more teams are building an abstraction layer for portability, guardrails, and cost/perf tuning. If it reduces complexity for users and improves reliability, you’re aligned with where things are going.

 

Thanks,