Agent Evaluation Dashboard — Full-Stack LLM Observability Tool

Teams running AI agents in production need visibility into how those agents actually perform not just whether they respond, but how accurate, fast, and cost-efficient they are. I built a full-stack observability dashboard to solve exactly that: a single place to track accuracy, latency, API cost, and schema compliance across multiple LLM models and agent types.

What I built

A responsive Next.js (App Router) frontend styled with Tailwind CSS — deep purple sidebar, live accuracy/latency charts (Recharts), filterable eval-run tables, and per-model performance comparisons

A Python FastAPI backend serving the dashboard's data layer, with token-based authentication protecting every route

Full session handling: login, protected pages, automatic redirect for unauthenticated users

A fully responsive layout collapsible drawer sidebar on mobile, reflowing grid on tablet/desktop

We've been building something for a while, and it's finally time to share it. Meet Zupa, an AI tool that helps you write LinkedIn and social media posts that actually sound like you. Most people know they should post more. The hard part is sitting down and turning a rough idea into something worth reading. Zupa takes that idea and shapes it into a post built around what makes people stop scrolling and engage. It's made for founders, marketers, agencies, and anyone building a personal brand who doesn't have hours to spend writing. Here's a quick look at what we've built 👇 We're opening early access now, and it's free to try. If you want a look, check out https://www.zupapost.ai/ Honest feedback is welcome too. That's how we'll make it better.

Really like the idea of using psychology to help people rewrite their posts! The UI feels thoughtful and fits the concept really well. It’s a focused tool I could see being useful in everyday writing, especially for making posts more engaging. How long did it take you to go from the initial idea to the finished website?