SODAX SDKDigital asset flows you can build and deploy with AI.
Promoted
Interview Copilot
Hears your interviewer and answers in under a second
14 followers
Hears your interviewer and answers in under a second
14 followers
Most interview tools are prep tools: mock questions before the call. Interview Copilot works during the real one. It listens to the interviewer through system audio - not your microphone - and answers technical questions in under a second. For live coding, screenshot the problem and get a working solution with the approach and time complexity, so you can explain the reasoning out loud. Native overlay: invisible on Zoom, Meet and Teams screen share. 16 languages. Free: 3 minutes + 2 screenshots
You've solved this pattern ten times. Under a timer, it's gone.
Live coding is a memory test disguised as an engineering test. The problem is on their screen, the clock is running, and the approach you know perfectly refuses to surface.
I added screenshot solving to Interview Copilot for exactly that.
⌘⇧S on the problem. A few seconds later the overlay has a working solution — with the approach and the time complexity, not just code.
That distinction was the whole design decision.
Handing someone 20 lines of Python doesn't help in an interview. You still have to explain why prefix sums, why a hash map, why O(n) and not O(n²). If you can't narrate the reasoning, pasted code sinks you faster than silence.
So the model is instructed to lead with the idea and the trade-off. The code is the footnote.
Three things that turned out harder than the AI part:
→ reading a problem off a screenshot reliably — shared editors, dark themes, half-scrolled text
→ keeping it under a few seconds, because after ~10 the moment has passed
→ making it a native overlay, invisible when you share your screen
Free to try — 3 minutes of recognition and 5 screenshot solutions, no card.
Hey Product Hunt 👋
I built Interview Copilot after bombing an interview where I actually knew the
answer. The question landed, my mind went blank for four seconds, and I never
recovered. The knowledge was there. The recall, under pressure, was not.
So I built the thing I wanted in that moment: a desktop overlay that listens to
the interviewer and streams a concise, structured answer onto your screen while
they're still talking.
Three things turned out to be much harder than "call an LLM":
⚡ Latency is the entire product. An answer that lands in four seconds is
useless — the silence already happened. Streaming STT + streaming output + firing
the moment the question ends (not when you press a key) puts the first words on
screen in under a second.
🎯 Detecting the END of a question is harder than answering it. People pause
mid-thought. Cut too early and you answer half a question; wait too long and the
moment is gone. That gate took more iterations than the AI.
🧠 I had to delete a feature to make it work. There was a cheap classifier in
front of the answering model deciding "is this worth answering?" It was perfect
on opening questions and completely silent on follow-ups — because "but why not
just use Kafka?" means nothing in isolation. It only makes sense against what
came before, and the classifier read it alone. Killing that gate is what made
the product usable in a real conversation.
Also in there: screenshot a coding task and get working code with complexity
analysis, resume-aware answers, and 16 languages — because interviewing in your
second language is its own kind of hard.
On the obvious question: yes, this works during a live interview, and no, I'm
not going to pretend otherwise. I think the interesting debate isn't "is this
cheating" but "what are we actually testing?" If a whiteboard round measures
recall under artificial pressure rather than the ability to do the job, tools
like this are a symptom, not the disease. I'd genuinely like to hear where you
land on that — including if you think I'm wrong.
Free tier, no card. macOS + Windows.
What would you want it to handle that it doesn't yet?
Pazi
The hard engineering here is impressive - capturing system audio without touching the mic and gating responses until a full thought lands is no small feat. Congrats on shipping it! 🚀
@zvonimir_sabljic1 Thanks a lot! Really appreciate it! 🚀