Navid Mani

Make your life fun!

[Navid Mani @NavidMani]

@f_j_j_ Thanks for showing that. Using simulations to catch confusing questions before real interviews makes sense. Have they caught anything you then changed in the interview?

[Navid Mani @NavidMani]

@telnyx @konnectile A useful next step could be asking attendees where booking felt confusing and what they'd change after the event. We're building Underheard for that kind of voice feedback. Is post-event research on Konnectile's roadmap? https://t.co/x6d2fevxjm

[Navid Mani @NavidMani]

@Speechmatics @livekit @aquavoice For the research interviews we're building at Underheard, preserving meaning matters: a dropped 'not' can flip a finding. What did the panel think today's voice benchmarks miss most? https://t.co/x6d2fevxjm

[Navid Mani @NavidMani]

@joinaboard @ultravox_dot_ai How do voice entries get compared on the leaderboard: community votes, task tests, or both? For interview agents, I'd want examples of recovery after an interruption or a corrected answer alongside any latency number.

[Navid Mani @NavidMani]

@Sadaqath_Umair I'd start with the reason to speak. Research interviews are one use case: letting people explain an experience in their own words, with room for follow-up questions. That's what we're building with Underheard: https://t.co/x6d2fevxjm

[Navid Mani @NavidMani]

@UsingOpen I'd try it for open-ended interviews, with a visible transcript and an easy switch to text if speaking isn't convenient. That's the research use case we're exploring at Underheard. How are you testing interruptions? https://t.co/x6d2fevxjm

[Navid Mani @NavidMani]

@TencentRTC We're working on research interviews at Underheard. For interrupted replies, does the integration expose what audio actually played, so an app can keep its conversation record aligned with what the participant heard?

[Navid Mani @NavidMani]

@stefanobertolix I'd interview customers about their last confusing order or return before picking a job. The answers could show whether they need information or a human-approved exception. That's our voice-interview use case at Underheard: https://t.co/x6d2feuZtO

[Navid Mani @NavidMani]

@Varindernagra Did you measure the pause from the caller's last sound or from the system deciding the turn was complete? For interview-style voice agents, I'd want to shorten the wait without cutting off someone who's still thinking.

[Navid Mani @NavidMani]

@HimeshPatel2121 This is a great reminder to make the audio chain visible. A tiny input-level meter plus a short record/playback check before a voice session would save so much guesswork. Did the OS pick a different mic, or was the stream silent?

[Navid Mani @NavidMani]

@bolna_dev @xan_ps @maitreya_wagh Our Underheard work is in voice interviews, so I'd love to see demos handle mid-sentence language switches and corrections. Will Mehfil let visitors report where a conversation breaks down, as well as what sounded good?

[Navid Mani @NavidMani]

@roopakdev28 Nice, what’s your agent’s job? We’re tackling research interviews with Underheard: asking people questions by voice and bringing their feedback together. Launched today if you want a concrete example: https://t.co/x6d2feuZtO

[Navid Mani @NavidMani]

@SuperBryn_AI @nikkithashanker We're building voice interviews at Underheard, so the move from a convincing demo to dependable calls is very relevant. In Pramod's experience, what broke first: understanding the caller, turn-taking, or completing the action?

[Navid Mani @NavidMani]

@thomasfogarasy For voice products, speed says little about whether someone feels heard. At Underheard we're building research interviews, and I'd use saved time to review confusing turns with real people. What should a quality check include beyond task completion? https://t.co/x6d2feuZtO

[Navid Mani @NavidMani]

@TheOGleonard Yes. A useful test is whether an interview changes the next product decision, not just fills a notes folder. We're working on the voice-interview side at Underheard; which research skill do you see builders skipping most? https://t.co/x6d2feuZtO

[Navid Mani @NavidMani]

@marias_martin Number 3 is our corner of the chaos 🙂 We launched Underheard today: voice interviews for customer research. Curious how you decide which questions deserve an interview when your PM day is this full: https://t.co/x6d2feuZtO

[Navid Mani @NavidMani]

@ktusing34 Congrats on the grant, Kyle! For Aria, I'd be curious about the awkward calls: a caller changing their mind, interrupting, or asking for a person. Is there a way to try those handoffs in the demo?

[Navid Mani @NavidMani]

@codewprince I'd add an explicit path for 'I don't know' and corrections to details already collected. A valid procedure can still trap a caller in a loop. How do you test those branches before a client call?

[Navid Mani @NavidMani]

@rokascode Picking thresholds before the sweep feels crucial here. I'd keep a separate set of old failure cases so a better overall score can't hide a broken call path. Did the regression appear in the test set, or only in live calls?