CrackIt

Overview

CrackIt is an AI interview preparation platform where candidates practise out loud. You set a target role and job context, then talk your way through a realistic interview — the system listens, transcribes, asks the next question in context, and scores the whole transcript at the end. It won the Nebius Everyday AI Hackathon 2025, hosted by Nebius & Sprint.dev.

Key Features

• Hands-free full-duplex voice loop over Socket.IO, using a Web Audio AnalyserNode for voice-activity detection that auto-commits each turn after 1.5s of silence, then streams it through Whisper-large-v3 and Llama 3 inference.
• Low-latency speech synthesis — Deepgram Aura speech synthesised as sentence chunks at 2-way concurrency, reassembling out-of-order arrivals in an indexed queue with head-of-line sequencing, skip-on-error recovery and a late-chunk grace window.
• Automated assessment engine on Nebius AI Studio, scoring each transcript with Llama-3.3-70B in strict JSON mode to return an overall rating, 4 competency scores, and structured strengths, improvements and actionable summary feedback.
• Full-stack platform — a React 18, Vite and Tailwind dashboard built on shadcn/Radix primitives, a Node.js/Express REST API, and Appwrite auth with per-user interview transcript persistence behind protected app routes.

Impact

Interview practice becomes repeatable and measurable instead of imagined. Candidates get a genuine back-and-forth conversation rather than a form, immediate structured feedback rather than a vague sense of how it went, and a history of sessions to track improvement against.