Mirak

AI can write the code. Can the candidate defend it? Take-home tests now measure little: any AI tool can produce code that passes every test. What still matters is whether a person understands the code they hand in. Defendable turns a job offer into a practical coding assessment with hidden tests, proven fair: a reference solution must pass them, and a flawed variant must fail each hidden test. The candidate builds it with any AI tool. Then Oracle, an AI interviewer, talks with them out loud about their own code. The recruiter gets a report that keeps measured facts apart from AI interpretations. Every quote and line of code it cites is checked, and each claim links to its moment in the recorded replay. No score: a person decides. For hiring, coding exams and school projects. Built on Meta's models: Muse Spark, Muse Wonder, Muse Voice Transcribe and MMS-TTS. Before the hackathon: the concept, the UI and a first backend. Built this week: the AI steps, the voice interview, verified tests, the report and the replay.