Algora

Algora is a competitive-programming practice platform: a browser IDE built on Monaco where you solve authored problems — or free-form playground code — in C++, Python and JavaScript, submitted to a judge that runs them against hidden test cases and reports verdict, runtime and peak memory.

Around that core it grows a whole ecosystem: user-authored problems with hidden tests, per-problem leaderboards, an AI tutor that knows your code and your last run, and live WebRTC rooms for pairing on a problem together.

Submissions never block the API. A POST writes a Processing row, runs a static complexity estimate, then pushes a JSON job onto a Redis queue. A decoupled worker pops jobs, executes the code inside an ephemeral Docker sandbox, and updates the row with per-test verdicts plus runtime and memory stats while the frontend polls for results.

Leaderboards rank each user's best submission by estimated time complexity first, then raw runtime — so an O(n log n) solution beats a faster-on-tiny-inputs O(n²) one.

Every run is a throwaway container: no network, 256 MB memory with swap pinned, one CPU, a 64-process PID limit, all capabilities dropped and no-new-privileges set. Solution runs go further — read-only root filesystem, a 16 MB tmpfs, and execution as the nobody user with the code mounted read-only.

Time limits are enforced twice: an in-container kill timer plus a host-side backstop. Out-of-memory kills are disambiguated from time-limit kills by checking wall-clock on exit code 137.

An AI tutor endpoint proxies to Gemini with a system prompt embedding the problem statement, your current code and the last run's summary — hints, not answers.

Meetings are full-mesh WebRTC: camera, mic and screen-share negotiated peer-to-peer through a WebSocket signaling relay, with glare avoided by making existing peers always initiate offers. Media never touches the server.