AjzDf&L#tZVXAjzDf&L#tZVXnAQm*eucK846nAQm*eucK846sjzDf&L#tZVXsjzDf&L#tZVXhcs*Y6E4mSOQhcs*Y6E4mSOQSVl7Rz+xfLHJSVl7Rz+xfLHJijzDf&L#tZVXijzDf&L#tZVXnAQm*eucK846nAQm*eucK846gVl7Rz+xfLHJgVl7Rz+xfLHJhcs*Y6E4mSOQhcs*Y6E4mSOQClick a letter to spin · double-click to replay

Students and early-career engineers struggle to map skills → roles → learning paths without expensive coaching.

Gemini-powered path builder + resume analyzer on a serverless Next.js stack (Drizzle + Neon) with structured outputs and actionable plans.

Shipped a full product loop: upload resume → analysis → personalized path — live in production.

Typing tests are single-player — there is no fair way to race someone live, because keystroke timing has to stay in sync across every client in the room.

Express 5 + Socket.IO rooms broadcast per-keystroke progress so every racer sees live positions; a custom engine computes WPM/accuracy on the client and Mongoose persists each result into a global leaderboard.

Room-code races with live opponent progress, a ready-up/rematch flow, and a persistent leaderboard — Next.js client on Vercel, socket server on Render, MongoDB behind it.

Agents and chat UIs can search, but results come back as unstructured text — you can see what was found, never how the sources relate to each other.

A Python MCP server exposes search, fetch, research and summarise tools over a FastAPI layer; the Next.js client turns the returned research graph into an interactive mind map.

A Dockerised MCP server wired to a web client that returns a navigable mind map instead of a wall of text — live in production.