Ask when each value is known
Don’t just drop features and retrain. For each top feature, check when it’s recorded compared with the moment of prediction.
An AI interviewer asks you real interview questions, digs into your reasoning, and rates every answer, so you know exactly what to fix before your real interview.
1 question free, no sign-up. 2 more with a free account. No card needed.
Your RAG system suddenly starts giving incorrect answers. What’s the first thing you investigate, and how would you prove that’s actually the root cause?
Covers it wellPartlyBarelyIllustrative: our view of typical coverage.
304 questions across 20 categories, all written by hand for data science interviews. Browse them by category
A product manager asks what "p = 0.03" means on your A/B test readout. How do you explain it, and what doesn't it tell them?
Each question is a full mock interview, and your first three are free. Paid plans have no question limit, within fair use; they differ only in how long you get.
Three questions, each a full mock interview, to see how it works. No card needed.
Two weeks of focused practice before an interview.
Steady practice for your job hunt. Cancel when you get the offer.
Three months for a full job hunt, with nothing to cancel.
Try up to 3 questions on your plan. If it’s not for you, ask within 7 days and we’ll refund you in full. Your free questions don’t count.
There’s no set number of questions. There are fair-use limits on AI use, because every turn runs on a paid AI model: typically around 30 questions in a day, 80 in a week or 200 in a month, depending on how long each conversation runs. They’re there to stop automated misuse, and they’re well above what most people practicing for an interview do. If you reach one, it frees up again as the day, week or month rolls on, and a question you’re in the middle of can always be finished.
You can, and it’s a decent start. The difference is what’s around the conversation. A general chatbot makes up its questions on the spot, tends to agree with you, and hands over the answer as soon as you’re stuck. Pondly’s questions are written by hand. The interviewer is built to follow up on what you actually said, and to give hints in steps rather than the answer. You finish every question with a written debrief: four ratings tied to your own words, and what to study next. You also don’t need to write a prompt or keep yourself on track.
If your interviews still have a coding round, keep LeetCode or StrataScratch for that. Pondly is for the rest of the interview: the rounds where you explain your thinking, like concepts, system design, troubleshooting and product sense. They work well together: drill the code there, practise talking it through here.
Yes, any time. You keep access until the end of the month you've paid for.
Yes. On a paid plan, try up to 3 questions, and if it's not for you, ask within 7 days of paying and we'll refund that payment in full. Just send us a ticket from the Support page. If something's wrong with Pondly itself, you're covered whatever you've used.
People landing a tech job, moving industries, or testing what they know — data scientists, analysts, ML engineers and AI engineers. The judgment rounds: concepts, system design, troubleshooting and product sense.