[radug14]

Hi everyone, I’m Radu.

In my personal experience, screening software engineers has always put pressure on internal engineering teams. Over the years, I’ve tried different approaches to streamline the process, but nothing has really fixed the problem of investing engineering time into screening.

At the start of the year I went through BetterStack’s recruitment process. Their first stage, an in-house built async screening test, was a revelation for me. I thought this was a fantastic alternative for an early stage in the recruitment pipeline. Back in February, while I was actively hiring at the startup I was working with, I prototyped a solution and trialed it - it was a success.

Fast forward a few months and I’ve now been able to turn the early prototype into a product. Meet Niju.

Niju replaces the traditional screening call with a 20-minute, asynchronous, recorded coding session.

A candidate gets a link, shares their screen, and "thinks aloud" while solving a practical coding challenge (no abstract algorithms).

After 20 minutes, Niju analyses the entire session: the code, the audio, and the thought process. It gives the hiring manager a concise report, transcripts, code playback and the raw footage with the important parts annotated. This means that, on average, a Niju interview takes 5 minutes to review.

* Cheating: Yes, a candidate can use Google. That's the point. I want to see how they solve a problem, not what they've memorised. The screen recording shows their whole process.

* AI: The AI does not produce a "pass/fail" decision. It just summarises the data to help a human make a better, faster, and more consistent decision.

* Stack: As a solo builder, I'm keeping it simple: SvelteKit, DrizzleORM, BullMQ, Postgres, Redis, Azure OpenAI.

The goal is to help busy engineering teams reclaim their time.

You can try the first interview for free.

I’ll be here for a while to answer questions and I'd be honored to get your feedback.

Radu

[abtinf]

I hope your product sees some adoption. Use of such a system would be a high-signal indicator for companies to avoid.

[radug14]

May I ask why?

I would never approve the use of async interviews further down the pipeline, but for screening purposes (from a candidate POV) I personally don't have any problems.

Keen to hear your point of view!

[radug14]

This is a super tough, but incredibly valuable thread. Thank you all for the raw feedback.

I need to be clear: I 100% agree with the core sentiment here. As a candidate, the hiring process is often broken, dehumanising, and feels like a one-way street. Many of you are right when saying a tool like this could be abused.

I'm not trying to automate the human part of hiring. I'm trying to fix the part that's already broken.

The real-world alternative at most companies isn't a friendly 1-hour chat with a senior engineer for every single applicant, that just doesn’t scale.

The alternative is:

1. A harsh, biased CV filter that rejects 95% of applicants in a couple of seconds.

2. A 4-hour take-home exercise that massively wastes your time and is genuinely pointless because anyone can vibe code it.

3. An algorithm test from a platform like HackerRank for which the majority of engineers have to prep many hours.

I built Niju to be less painful than those. It's a 20-minute, practical, "think-aloud" test. The AI's only job is to summarise the data so a human can review it faster, making it more likely they'll widen the funnel and give more people a shot beyond just their CV.

My goal isn't to replace engineers but to stop wasting their time on a broken process, so they can have better, human interviews with the top 20% of candidates.

It's a massive challenge, and this thread, as well as most of the others, show the raw nerve I've hit.

[UncleMeat]

> I'm trying to fix the part that's already broken.

The broken part appears to be that somebody spends 20 minutes screening a candidate rather than 5 minutes reviewing an AI summary of the screen. Inefficient, maybe.

Why does this actually get rid of the alternatives? You've reduced the active time to screen a candidate by 75%. But a CV filter reduces it by 95%. Why would your system involve less prep than HackerRank? It is still a test with a critical outcome. The idea that if only we didn't ask algorithmic questions that nobody would spend time rehearsing for interviews is ridiculous.