Promptly

Inspiration

The future of programming is vibe coding, but it comes with tradeoffs. If the assistant takes over, you lose track of the code; if you avoid it, you lose speed. I built Promptly because I wanted a place to practice finding that balance between not enough and too much

What it does

Promptly is an isolated training environment where developers solve programming challenges with AI. You pick a category (React, Python, C++), work in a guided sandbox, and submit your solution. Each submission is built in its own container, logs are captured, and an AI analyzer reviews the build output and chat history to give targeted feedback on code quality, functionality, production readiness, and token efficiency.

How I built it

Next.js 15 + React 19 frontend with Tailwind and Sandpack for the in-browser coding environment.
tRPC + Drizzle ORM + Postgres for typed APIs, submissions, and analytics.
Go-based judge server that spins up isolated builds, captures logs, and exposes an A2A/MCP-based analyzer agent.
Kubernetes + BuildKit for containerized, sandboxed builds.

Challenges I ran into

Orchestrating sandboxed builds safely while still streaming usable logs back to the client.
Coordinating the judge server’s A2A agent flow with the Next.js submission pipeline.
Self hosting a deploying agent A2A server

Accomplishments that we’re proud of

A full end-to-end loop: solve → submit → isolated build → AI analysis → feedback dashboard.
A working sandboxed environment that mirrors real dev workflows.