LogicForge - AI-Proof Technical Evaluation

Take-home coding assignments and multiple-choice quizzes are trivial to outsource to an LLM, so they've stopped signaling engineering intuition. LogicForge needed a format that scores how a candidate reasons under time pressure — debugging, tracing, and predicting code behavior — rather than what code they can generate.

• 01Turborepo monorepo splitting the Next.js client, API gateway, WebSocket game services, question engine, anti-cheat scoring, and a Go code runner into independently deployable packages sharing typed contracts.
• 02WebSocket game services keep each candidate's session state in Redis so scoring and timers survive service restarts without dropping an in-progress round.
• 03A separate Go code runner isolates untrusted candidate code execution from the Node.js services, keeping the blast radius of a malicious submission contained to a disposable sandbox.
• 04PostgreSQL holds structured evaluation/session data; MongoDB stores the less-structured question bank and anti-cheat event logs, split by access pattern rather than forcing one database to do both jobs.

Designed the monorepo boundaries and the WebSocket session-state model, and built the anti-cheat scoring pipeline that flags suspicious answer timing/patterns during a live round.