wa5ina

A curated directory of documented incidents where AI-generated and vibe-coded software failed in production.

I got tired of paying $35/mo just to see company-specific tags, so I scraped 1,500+ recent questions across 458 companies to map their actual statistical biases.

I ran some basic frequency math. For example, looking at 100 recent Apple questions, their difficulty is standard (15% hard), but they test BSTs at 4.5x the industry average and backtracking at 2.5x.

The site is completely free, no login required. Just built it to make interview prep slightly less of a black box. Happy to answer any questions on how I scraped/categorized the data.

Hey HN, OP here. I built this because grinding raw LeetCode only teaches you how to pass test cases, not how to communicate. The AI interviewer is specifically prompted to interrupt you and ask 'why' if you start coding without explaining your approach first—just like a real senior engineer would. I'm still tuning how aggressive the interruptions are. Let me know if the pacing feels natural or annoying

on point, this should in the roadmap

Thanks for the feedback! my justification for the pricing, are backed by the fact that once you purchase a package, you get lifetime access to the problems database, which updates in real time. specifically, we have a cron job that mines data every hour for 24h, which is free but (at least for now), but LLM processing isn’t, so the justification is that u get continuous value after u made one purchase, and your credits are only used for generating reports. u can always add more credits as needed. I’m curious though, does this make sense to you? If not, what price point would feel fair or approachable for testing it out?

> “What is validated problem"?

The way I see it at least is, a validated problem is one that enough people face, frequent and cause a certain level of frustration, currently each problem in broblems have a “frustration level” extracted from language used which serve “problem” and “frustration level”, but “enough people” is not yet in the equation, it’s something i wanted so badly to add, but decided not to in MVP, is to aggregate group of people that faces the same problem, and be able to browse and analyse that in a centralised location , i think that’s valuable , and i see me using that feature frequently, will try to add this feature asap. but what in ur opinion is a “validated problem”.

> “did someone verify that the people having the problem would actually pay for a solution,” ?

Great question, my simple way currently is that each problem have a "context score" that decide to what level this problem is a “productizable problem”, "renting about an analytics tool" have a higher context score while "existential rent about existence" have a lower one, this is simply done by prompting the LLM, only things that above 8 are considered actionable problems.

> “do the reports track when problems were identified and seek out whether solutions have been created since the complaints arose”

the dive-in reports currently show existing players in the space, but i don’t yet track problem timelines or follow up on whether solutions have been created since. sounds like a solid idea, but i guess it’s a bit beyond the scope of this initial version.