Vapi's 88-Person Title Pool: Source Voice-AI Engineers by Skill, Not Title
Vapi just raised $50M and Ring routes 100% of calls through it. Here's the skill-based sourcing map for voice-AI engineers before Series B closes.
On May 12, 2026, Vapi closed a $50M Series B led by Peak XV at a roughly $500M post-money valuation, with Amazon Ring routing 100% of its inbound customer calls through the platform. The new capital is going into engineering and go-to-market, which means the next 90 days will absorb a sizeable chunk of the senior real-time audio and telephony talent that exists in the US. If you're hiring against this category, the bad news is that the obvious title pool is essentially empty. The good news is that the skill graph is unusually legible.
The title pool is a trap
Search LinkedIn for current titles like "Voice AI Engineer," "Speech Engineer," "Real-time Audio Engineer," "Telephony Engineer," and "Conversational AI Engineer," and you will find roughly 88 people in the US. Twenty of those are legacy telephony titles at Apple Retail, ScanSource, AT&T Interactive, Walt Disney, and Wheels Up. The category Vapi just validated does not yet have an accepted job title.
That's the entire title-based pool, and Vapi alone wants to grow its ~100-person team meaningfully off a $50M round. Sierra, Decagon, PolyAI, Bland, Retell, and ElevenLabs are pulling from the same well, plus Cartesia and Deepgram on the model side. If your sourcing strategy starts with a title filter, you've already lost.
The pool that actually matters is skill-based. Around 5,600 senior+ US profiles list WebRTC or SIP. That cluster skews to legacy telecom (Genesys, Patton Electronics, Nokia Siemens Networks, Bell Labs alumni), which is the opposite of what most recruiters assume when they hear "AI voice agent recruiting." The hard part of this stack isn't the model. It's the telephony.
What Ring's "two weeks to production" actually tells you
Ring's VP of Software Development, Jason Mitura, said on the record: "We went from zero to production in two weeks, and 100% of our inbound volume now runs through Vapi." Ring evaluated more than 40 vendors before picking Vapi. Read that quote as a hiring spec, not a press line. The engineers who make a two-week enterprise telephony cutover possible are not research scientists. They are integration-focused, reliability-obsessed, on-call-comfortable systems engineers who happen to know how LLMs fail.
The hard part of voice AI is not the LLM. It is bridging PSTN to WebRTC without dropping the call. </pull> That profile sits at the intersection of four skills that rarely cluster on one resume: 1. **Real-time audio transport.** WebRTC, RTP, jitter buffers, echo cancellation, codecs (Opus, G.711). 2. **Telephony interop.** SIP, SDP, PSTN gateways, DTMF, carrier quirks. Often learned the hard way at Twilio, Vonage, Telnyx, Plivo, or Bandwidth. 3. **Low-latency inference orchestration.** Sub-500ms budgets at scale, streaming STT/TTS, VAD tuning, barge-in handling. 4. **LLM tool-use plumbing.** Function calling with evals, guardrails, rollback, and the kind of paranoia you only develop after a model has confidently transferred a customer to the wrong department. Co-founder Jordan Dearsley described the job as "taking this indeterminate beast that is a model and taming it." That's the skill premium right now. Pretty TTS demos are commodity. Tool-use reliability is not. ## The competitor alumni map TechCrunch's own coverage names the direct competitor set. Treat it as your primary sourcing list: - **Sierra, Decagon, PolyAI, Bland, Retell** for full-stack voice agent builders. - **ElevenLabs, Cartesia, Deepgram** for streaming model and audio infrastructure engineers. Then layer in the telephony-native adjacencies, which are where the genuinely scarce skill lives: - **Twilio** (especially Flex and Programmable Voice teams), **Vonage**, **Telnyx**, **Plivo**, **Bandwidth**. - **Daily.co** (the team that built Pipecat) and **LiveKit** itself. - **Genesys, Five9, NICE, Cisco BroadSoft** for the contact-center veterans who already know what enterprise call routing demands. The non-obvious move is the contact-center incumbents. A senior engineer at Five9 who has spent five years thinking about call deflection, IVR fallback, and queueing is closer to "voice AI engineer" than a generalist LLM engineer who has never debugged a SIP INVITE. The retraining gap from telecom to LLM tool-use is smaller than the gap from LLM to telecom. ## Why GitHub matters more than LinkedIn here The Pipecat and LiveKit ecosystems are the most efficient sourcing channels nobody is naming. LiveKit is open-source and one of the most widely deployed WebRTC media servers in the world. Pipecat, built by Daily, has become the de facto Python framework for realtime voice/video AI. Their GitHub contributor graphs, issue threads, and Discord channels are a public, free, candidate index that bypasses the LinkedIn arms race entirely. A few concrete starting points: - **livekit/agents** on GitHub. Filter contributors by commits in the last 12 months. - **pipecat-ai/pipecat** and **pipecat-ai/pipecat-flows**. The flows repo is newer and the contributor set is smaller and more senior. - The OpenAI Realtime API early-access cohort and anyone shipping demos against Amazon Bedrock AgentCore or Google Vertex ADK. LinkedIn's keyword search will not surface these people cleanly because their day-job titles are "Senior Software Engineer" at companies you've never connected to voice. This is exactly the friction we built [Refolk](/) for: you describe the person in plain English ("senior engineers who contribute to LiveKit or Pipecat and have shipped SIP integration in production"), and you get a ranked shortlist that joins GitHub activity, LinkedIn history, and open-web signals into one profile. Title-based sourcing cannot find these people. Skill-and-signal-based sourcing can.
refolk prompt: Find US-based senior engineers who contribute to LiveKit or Pipecat on GitHub, list WebRTC or SIP on LinkedIn, and have worked at Twilio, Vonage, Telnyx, Daily, or a contact-center vendor in the last 5 years. note: You'll get a ranked shortlist that the title search "Voice AI Engineer" literally cannot return, because the title barely exists. slug: jry82yt6yq
## The latency filter is a real screening tool
Vapi's product page advertises sub-500ms latency at scale. The community benchmark for Pipecat on Daily averages 800 to 950ms end-to-end, with LiveKit holding a "slight edge" via tighter transport integration. Those numbers are not trivia. They are an interview rubric.
Ask a candidate to walk you through where milliseconds go in a voice agent loop: mic capture, VAD trigger, STT streaming, LLM first-token, TTS first-audio, network jitter, playback buffer. An engineer who can decompose that budget without notes has done the work. One who hand-waves to "we use a fast model" has not. This is a cheap, fair, technically honest screen, and it separates the 5,600-strong WebRTC/SIP pool from the much smaller pool that has actually shipped a sub-second voice agent.
## Vapi's own developer community is a pool
Peak XV's Arnav Sahu compared Vapi to "Supabase, PostHog, Better Auth and ClickHouse." That's an explicit bottom-up, developer-led thesis. Vapi has cited a developer community in the seven figures and has handled more than a billion calls cumulatively, currently processing 1 million to 5 million calls per day. Enterprise revenue is up 10x since early 2025.
```stat
number: 10x
label: Vapi's enterprise business growth since early 2025
note: That's the velocity your offer letter is competing against.
</stat>
The implication for sourcing: the people building on Vapi are themselves the pool. Anyone who has shipped a production Vapi assistant for an internal team has done, in miniature, the work Sierra and Decagon are hiring for. Same for builders on Bland, Retell, and the LiveKit Agents framework. These are usually 2-to-4-engineer teams inside larger companies, and they almost never have "voice" in their title.
Vapi's named reference customers are the warm pool: Intuit, New York Life, ServiceTitan, Instawork, Kavak, Cherry, UnityAI. Each has at least one internal practitioner who just integrated a voice-AI vendor against real enterprise constraints. Those people are not on any "AI engineer" sourcing list, and they are exactly the profile Mitura's "two weeks" comment describes.
## A concrete 30-day sourcing plan
If you're a founder or recruiter staffing against this category before Vapi's round actually deploys, here is the order of operations.
### Week 1: Map the GitHub graph
Pull the last 12 months of contributors to livekit/agents, pipecat-ai, and the major Vapi SDK repos. Cross-reference against LinkedIn for current employer, seniority, and location. This is the highest-signal, lowest-noise pool you will find, and almost nobody is mining it systematically. A plain-English query in Refolk against "active LiveKit Agents contributors in North America with 6+ years of backend experience" collapses what used to be a multi-day spreadsheet exercise into one search.
### Week 2: Work the telephony alumni list
Twilio, Vonage, Telnyx, Plivo, Bandwidth, Daily, Genesys, Five9, NICE. Filter for engineers who have shipped SIP, WebRTC, or contact-center integrations. The pitch writes itself: the skill you built over 8 years at Twilio is suddenly the rate-limiting input for the most-funded category in AI infrastructure.
### Week 3: Mine the reference customers
Intuit, ServiceTitan, New York Life, Instawork, Kavak. Look for engineers whose recent project history mentions voice agents, IVR replacement, or contact-center automation. These are battle-tested practitioners who already know what production looks like and who are usually paid like generalist platform engineers, which means the comp delta in your offer is meaningful.
### Week 4: Close on shipping velocity, not research credentials
The Ring deployment was two weeks zero-to-production. Your interview loop should reward engineers who have shipped reliability work (evals, rollbacks, on-call runbooks for nondeterministic systems) over engineers with the most polished demo. Dearsley's "taming the indeterminate beast" framing is the right cultural filter.
Across all four weeks, the constant is that the pipeline lives in skill stacks and open-source activity, not in titles. That's why "telephony engineer pipeline" beats "voice AI engineer sourcing" as a search strategy, and it's why a tool like Refolk earns its place in this workflow: plain-English queries join the GitHub, LinkedIn, and open-web signals that title-based search will never connect on its own.
The Series B closes once. The talent it pulls in is gone for the next two years. Move now.
## FAQ
### Why are there so few people with "Voice AI Engineer" as a current title?
The category only became mission-critical in the last 12 to 18 months, and most practitioners hold generalist backend or platform titles at their current employers. Internal data shows roughly 88 US profiles with adjacent current titles, and a quarter of those are legacy telephony roles unrelated to LLMs. Title-based sourcing for this niche will return a near-empty list. Skill-based and GitHub-based sourcing returns thousands.
### Which alumni pool gives the highest hit rate for voice-AI roles?
Telephony-native companies, not AI-native ones. Twilio, Vonage, Telnyx, Plivo, Bandwidth, and the contact-center incumbents (Genesys, Five9, NICE) have the SIP and WebRTC depth that turns out to be the rate-limiting skill. Pair that with anyone who has shipped LLM tool-use with evals and guardrails, and you have the profile Vapi, Sierra, Decagon, and Bland are all chasing.
### How do I evaluate voice-AI candidates without becoming a voice-AI expert myself?
Use the latency budget as a screening exercise. Ask the candidate to decompose where milliseconds go in a voice agent loop, from mic capture to TTS playback. Vapi advertises sub-500ms at scale; Pipecat on Daily benchmarks 800 to 950ms end-to-end. Engineers who can talk through those numbers without notes have done the work. Engineers who can't, haven't.
### Is Vapi's own developer community actually a sourceable pool?
Yes, and it is undervalued. Peak XV explicitly framed the investment thesis around bottom-up developer adoption, comparing Vapi to Supabase and ClickHouse. Builders who have shipped production Vapi assistants, especially inside named reference customers like Intuit, ServiceTitan, and New York Life, are doing exactly the work Sierra and Decagon are hiring for, but they almost never carry "voice" or "AI" in their title.