Komal P

Founder @ & | Building AI-powered CPaaS & CCaaS | Voice AI & Conversational AI

[Komal P @komalpvoiceai]

🔧 Hands-on experience in SIP traces, RTP/media issues, SBCs (AudioCodes/Anynode), carrier integration & complex voice troubleshooting is what we’re looking for.

📩 Send your resume to [email protected]

Know someone who fits? Repost! 🔁

#Hiring #VoIP #SIP #Telecom #VoiceEngineer

[Komal P @komalpvoiceai]

🚀 We’re Hiring! Voice Engineer (L2/L3) at https://t.co/aCFxNa5qzp

🔹 SIP / VoIP / SBC
🔹 Microsoft Teams Telephony
🔹 PSTN, SIP Trunks & CCaaS
🔹 Wireshark & PCAP troubleshooting
📍 GIFT City, Gandhinagar, Gujarat
🏢 Onsite | 💼 Full-time
💡 Experience: 4+ years
🧵👇

[Komal P @komalpvoiceai]

most of my time lately is number porting, usage-based billing edge cases, and call quality monitoring for dialog365 — not glamorous, but it's where things actually break. if you're building any of that, say hi. what are you building?

[Komal P @komalpvoiceai]

Hiring devops engineer with 1 to 3 years of experience. Fwd resume to [email protected]

[Komal P @komalpvoiceai]

I used to think teasing a feature early built anticipation. wrong — it just meant weeks of explaining delays to people who were waiting. now nothing gets posted until it's live and someone can actually use it.

[Komal P @komalpvoiceai]

unpopular opinion: chasing sub-300ms latency in voice agents is mostly wasted effort. people tolerate half a second of silence in normal conversation without noticing.

the money's better spent on turn-taking and interruption handling than shaving another 100ms off the pipe.

[Komal P @komalpvoiceai]

trying to decide whether siprtc should auto-failover to a backup STT provider on error, or just hard-fail and alert. auto-failover hides outages from users but if the backup is worse quality, they never know why calls got harder to transcribe. no clean answer yet.

[Komal P @komalpvoiceai]

I'm 42.

Solo builder from India.

Looking to connect with more builders & indie hackers!

[Komal P @komalpvoiceai]

trying to pick a default jitter buffer size for dialog365. 150ms feels safe on paper but I don't have real data on how many callers hit bad networks. tempted to ship it anyway and log drops to tune later.

[Komal P @komalpvoiceai]

i assumed more compute would fix latency spikes in voice pipelines. wrong — the bottleneck was queueing at the media server, not cpu. now i check queue depth before i ever add a box.

[Komal P @komalpvoiceai]

@markproduct AI voice agent with AI native call center https://t.co/O7TidTcb1x

[Komal P @komalpvoiceai]

@olegzanx The transfer is with context and summary of conversation so caller has not to repeat. And human agent has full context of caller.

[Komal P @komalpvoiceai]

a voice agent should know when to stop automating. dialog365 has AI-native handoff to a human contact centre agent, so the call can move to a person when the conversation needs one.

[Komal P @komalpvoiceai]

i’m building dialog365 for the part after the voice AI demo: low-latency conversation, a choice of STT and TTS providers, and a clean handoff to a human agent when automation is no longer useful.

[Komal P @komalpvoiceai]

dialog365 voice agents support elevenlabs, microsoft, google and deepgram for STT and TTS. contact centres can choose the speech stack that fits their calls instead of being tied to one provider.

[Komal P @komalpvoiceai]

@lightsilver323 https://t.co/RRhPZTtys9

[Komal P @komalpvoiceai]

@lightsilver323 https://t.co/aCFxNa5qzp

[Komal P @komalpvoiceai]

@ElevenLabs Music creation by ai agents is area to be explored

[Komal P @komalpvoiceai]

unpopular opinion: chasing sub-300ms latency in voice agents is mostly wasted effort. people tolerate half a second of silence in normal conversation without noticing.

the money's better spent on turn-taking and interruption handling than shaving another 100ms off the pipe.

[Komal P @komalpvoiceai]

trying to pick a default jitter buffer size for dialog365. 150ms feels safe on paper but I don't have real data on how many callers hit bad networks. tempted to ship it anyway and log drops to tune later.