PeerPrep Collaborative Interview Prep

Overview

A platform where two users are matched and dropped into a live coding session with a shared editor, audio, and a curated question bank purpose-built for technical interview practice.

Technical Challenges

Collaborative Editing with CRDTs

Used Yjs (a CRDT library) synced over y-websocket to handle concurrent edits. CRDTs were chosen over OT because they resolve conflicts without a central coordinator, letting each peer apply remote operations out-of-order and converge to the same state critical for a real-time coding environment where network jitter is common.

Peer Matching Service

Matching is based on difficulty preference and topic tags. The matching service maintains per-category queues and attempts exact matches with a timeout, after which it relaxes constraints (drops topic requirement, then difficulty). Matched pairs are notified via Socket.io and a session record is created atomically to prevent double-matching.

Microservices & Orchestration

Services (User, Question, Matching, Collaboration, History) run as separate containers orchestrated with Kubernetes. Inter-service communication uses HTTP for synchronous calls and Socket.io rooms for real-time events. k8s readiness probes prevent traffic from reaching pods before their WebSocket servers are ready.

Audio Communication

Peer-to-peer audio is handled over WebRTC with the collaboration service acting as a signaling relay. The signaling channel reuses the existing Socket.io connection to avoid maintaining a separate WebSocket.

Polyglot Persistence

User accounts and session history are stored in PostgreSQL for referential integrity. The question bank uses MongoDB to accommodate flexible metadata schemas across different question types.