Probe - Realistic coding interviews, with AI use encouraged

Hey Product Hunt 👋 I'm Paran. Berkeley CS, ex-Citadel quant, and the maker of InterviewDen (free AI mock interviews, my other product). InterviewDen helps people get ready for interviews. Probe came from the other side of the table. For a while I was running six one-hour technical interviews a week and rejecting about 95% of the people I talked to. The candidates weren't the problem. The problem was that I had no way to tighten the funnel at the top, so every weak applicant still cost a senior engineer a full hour on a call that was going nowhere. And the whole time, every candidate had an AI assistant open in another tab. The interview just pretended that tab didn't exist. So we ban the tools, fall back to whiteboard recall or three-week take-homes, and then act surprised when the signal is bad. (In one study of ~19,000 interviews, 38% of candidates tripped AI-cheating flags anyway.) Probe makes the opposite bet: assume the AI is there, and measure judgment instead. What it does: 👁️ A silent "watcher" AI. While the candidate works, a second AI watches every prompt, edit, and test run. It never speaks to them. It just grades the work against your rubric, the way a senior engineer would if they were reading over a shoulder. 📎 Every score is cited to the transcript. No black-box "7/10." You see the exact moment behind each dimension, and you can override the recommendation in one click. 🧩 Real tasks, not trivia. Five formats (productionize, build-a-feature, refactor, review-an-AI-PR, open-ended build) in Python, Java, or C++. No LeetCode, no whiteboard. 🎙️ Optional voice debrief. After they submit, a voice AI asks them to walk through what they built, so you know they actually understand the code and didn't just paste what the assistant generated. ⚡ Async, no install, scorecard in minutes. You send a link. No scheduling, no reviewer queue, no "we'll get back to you in three weeks." The part I'm proudest of: it tells "used the AI well" apart from "let the AI do it," and those look nothing alike in the transcript. The people who outsource their thinking leave a trail of vague prompts, unverified pastes, and code they can't defend. The ones who can actually think show their work. Who it's for: engineering hiring teams, hiring managers, and founders running their own technical screens. Basically anyone drowning in take-home review, or anyone who's sat through that 95%-reject call and wanted the hour back. One honest note, because I'd rather you hear it from me: we're early. Python, Java, and C++ today (custom tasks from your own repo and more languages are coming), no ATS integrations yet, and the scorecard is a recommendation you can always override. What I'd love from PH: if you hire engineers, try Probe on a real role at your company. I'll help you set it up personally. And if you just want to feel it from the inside, run a session as if you were the candidate and read your own scorecard. Either way, come tell me what worked and what didn't. If you think the whole idea is wrong, I especially want to hear that too. Thanks for checking it out. AMA below 🙏 — Paran