Building realistic AI interviews is WAY harder than I expected
I thought the difficult part would be AI responses.
Turns out the hardest part is making the interaction feel human:
interruptions
pressure
recruiter skepticism
emotional reactions
trust changing during answers
follow-up questions based on weak points
Most interview tools feel like scripted Q&A.
I’m trying to build something that feels closer to:
“a recruiter who has actually read your CV and is judging your answers in real time.”
As a solo founder, I’ve also learned:
launching is not the hard part anymore.
Distribution is.
You can spend months building something technically impressive and still hear complete silence after launch.
Right now I’m focusing on:
improving realism
reducing onboarding friction
talking to users directly
posting consistently instead of waiting for a “perfect launch”
What made your product finally start getting traction?
It's possible it's just too soon in overall AI adoption curve for people to feel comfortable too. Don't beat yourself up.
That’s honestly something I’ve been thinking about too.
A lot of people are still in the phase of “AI can answer questions,” while realistic AI interaction is a much newer expectation psychologically.
I think trust and comfort with AI experiences may evolve gradually, especially for something as emotionally sensitive as interviews and career evaluation.
By the way, if you're open to it, I'd love honest feedback on what I'm building. WorkZo AI is currently in beta, and I've been focusing heavily on interview intelligence and realism rather than polish. If you get a chance to try it on Product Hunt, I'd genuinely appreciate any thoughts on what feels realistic, what doesn't, and where it could improve.
https://www.producthunt.com/products/workzo-ai
Latency is brutal for voice AI too — I've been building a voice agent for local businesses and the difference between 800ms and 1.5s response time completely changes whether it feels human or robotic. The STT → LLM → TTS pipeline needs to be really tight.
Completely agree. Even tiny delays change the entire emotional feel of the conversation.
Once responses cross a certain threshold, users stop “talking to someone” and start feeling like they’re waiting for a system. That illusion breaks very fast in voice AI.
Working on voice agents myself (launching September). The "talking to a system" threshold is real — anything above 800ms feels mechanical. Have you tested with sub-500ms TTFB? Curious if that flips the perception back.
I've found the perception changes dramatically below 1 second, but interestingly latency isn't the only factor.
Even with decent response times, things can still feel robotic if turn-taking, interruptions, pauses, and reactions aren't handled naturally.
In interview simulations especially, users seem to notice conversational realism as much as raw speed. Curious to see what you learn as you get closer to launch.
The "too perfect" break point is exactly what we're hitting with voice agents for small businesses. A dental clinic receptionist isn't polished — she says "mmm" and sometimes forgets things. If the AI sounds more professional than the human it's replacing, clients notice something's off.
The realism vs discouragement balance is brutal. We're testing intentional micro-imperfections. Early results are promising.
The calibration problem alone must be brutal - one second too robotic and candidates disengage, the next second too permissive and you're not actually evaluating anything useful. Curious what your biggest surprise was in terms of what actually trips these models up?
Honestly, one of the biggest surprises was how easily the interaction breaks if the AI becomes “too perfect.”
Real recruiters are messy:
If the AI sounds too polished or too linear, users immediately stop feeling pressure.
Another difficult part is balancing realism vs discouragement.
Too aggressive and users shut down.
Too soft and it stops feeling useful.
The strongest part here is the “recruiter who has actually read your CV” framing.
That is much more interesting than another AI interview-practice tool, because it shifts the product from generic Q&A into realistic career judgment: pressure, skepticism, follow-ups, weak-point detection, and trust changing during the conversation.
I’d build the positioning around that. People do not just want to rehearse answers. They want to know how their answers hold up when someone is actively evaluating them.
One thing to be careful with is making the product sound too much like a practice chatbot. If the experience is moving toward realistic interview simulation and professional readiness, the brand should probably feel more serious than a generic AI interview tool.
Beryxa. com would fit that direction well because it feels like a professional decision-training platform rather than a scripted mock interview app. The product sounds like it is selling confidence under pressure, so the name needs to carry trust before users even start the first session.
Really insightful point.
I’ve been realizing that users are often not looking for “practice” alone.
They want realistic evaluation under pressure.
The moment the recruiter feels like they’ve actually read your CV, noticed inconsistencies, questioned vague claims, and changed trust dynamically, the interaction becomes emotionally different from generic Q&A.
And yes — positioning is turning out to be almost as important as the technology itself. The wording completely changes how serious and credible the experience feels before users even enter the interview room.
Exactly. Once the product feels like realistic evaluation under pressure, the frame changes completely.
It is no longer “AI interview practice.” It becomes:
Can your answers survive skepticism, follow-ups, weak-point detection, and changing recruiter trust?
That is the line I would build the homepage and outreach around.
If useful, I can put together a quick positioning/outreach pack for this: sharper category angle, homepage hero line, 3 cold messages, 3 LinkedIn DMs, and follow-ups you can test with job seekers, career coaches, or bootcamp communities.
Not a long strategy doc. Just copy you can use immediately.
I’m doing a few at $49 to move fast.