Carbon Technology Group · BeeSuite AI · Remote · Contract-to-Hire
The short version
We've built a system of voice agents — over 9,000 production calls answered, more than 5,000 AI voice interviews held by our virtual recruiter, and several voice use cases nobody's built before in the works. Real customers, a waitlist to onboard, and one constraint on growth: how fast we can automate and manage scale. That's where you come in.
Now we're moving from Retell to LiveKit, training our own voice models, and scaling from thousands of calls a month to millions. We need one more engineer to own that — someone who refuses to be told what's possible.
What we believe
Magic is engineered. A caller feels heard because barge-in fires in 200 milliseconds, turn-taking never talks over an elderly caller, and every answer is grounded in her actual account. When the technology is done right, it disappears — what's left feels like someone competent and warm on the other end of the line.
An agent that sounds like a robot is a failure, even if the code is correct.
What you'll own
Voice, end to end. The Retell → LiveKit migration, then the platform every agent runs on — turn detection, interruption, SIP/WebRTC, multi-agent handoff, multi-tenant config. Migration succeeds when customers never notice. And it's not one receptionist: inbound, outbound, interviews, check-ins, surveys — a hive of voice agents, each pushing a different limit.
Past the APIs. Train and fine-tune our own voice models. Own inference and drive cost-per-conversation down as volume explodes. Ground agents in real customer data — Supabase, pgvector, retrieval into HiveMind, our intelligence layer that gets smarter with every call. Prove it with evals: "first-response accuracy went up 12% last week" — and show the data.
Wherever it takes you. Voice is the front door, not a cage. We keep the team tiny on purpose and automate everything else, so the people on it touch everything — backend, orchestration, product, whatever the mission needs. Find a better path than the one we planned? Take it. That's the job.
How we work
We're real coders. We were shipping long before AI wrote a line of code — but we embraced it early, and now one engineer here ships like five. If your instinct on a boring problem is "I'll build an agent so nobody ever does this again," you'll feel at home.
Ship fast. Remote-first, async-preferred, output over optics. No committees between you and production.
We run on EOS. One Team. Own It. Trajectory Wins. Reinvent What's Possible. Not wall art — how we decide.
Our stack
LiveKit Agents (Python) · Twilio + LiveKit SIP/WebRTC · multi-model LLMs (Anthropic, OpenAI, Google) · Deepgram / Cartesia / ElevenLabs today — our own voice models next · TypeScript + n8n orchestration · Next.js + Supabase (Postgres, pgvector, RLS) · HiveMind learning from every transcript
Who you are
You've shipped real-time voice AI to production and kept it alive at volume — LiveKit ideal, but if you built it on Retell, Vapi, Pipecat, or raw WebRTC pipelines, we care what you shipped, not which logo. Strong Python in streaming, async, real-time systems. You've built the full loop — STT, reasoning, TTS, VAD, turn-taking, tool calls — and debugged it at 2am.
Bonus points: eval frameworks for LLM systems, voice model fine-tuning, inference and cost optimization, multi-tenant SaaS, telephony scars.
Most of all: you read "that's not supported" as an invitation.
Structure: Contract-to-Hire
We're not asking you to quit your job and hope.
Phase 1 — 3-month paid contract, real production work from week one. Rate negotiated to fit your situation, benchmarked against the full-time band below. 20 hrs/week while transitioning is fine. Either side can end it — no hard feelings.
Phase 2 — Full-time conversion on mutual fit: $150–200K base + real equity as one of our earliest engineers, up to $20K annual bonus, health insurance, flexible PTO. Path to senior IC or technical leadership.
Exceptional candidate who needs to go straight to full-time? Let's talk, the structure serves the fit, not the other way around.
Deliberately funded. We raise carefully, and only what the mission needs, so the people building this keep real ownership and a real outcome. Long runway, no growth-at-all-costs pressure, and equity that's meant to matter.
Successful Founders with exits and ex-Apple leadership.
Skills Required
- Experience shipping real-time voice AI systems to production and operating them at scale
- Strong Python expertise in streaming, async, real-time systems
- Built the full voice stack: STT, LLM reasoning integration, TTS, VAD, turn-taking, and tool calls
- Experience with SIP/WebRTC and telephony platforms (Twilio) and LiveKit or equivalent real-time voice platforms
- Experience integrating and operating LLMs and multi-model stacks (Anthropic, OpenAI, Google)
- Experience with vector retrieval and databases (Postgres, Supabase, pgvector) and multi-tenant data concerns (RLS)
- Experience training and fine-tuning voice models, inference optimization, and building eval frameworks for LLM systems
- Familiarity with Deepgram, Cartesia, ElevenLabs or similar STT/TTS services
- Experience with TypeScript, Next.js, and orchestration tooling (n8n) for full-stack or orchestration work
- Startup experience and willingness to operate across backend, orchestration, product, and production support
