You'll be the person who builds the machinery behind our AI calling agents. At JoyzAI, that means the real-time voice pipeline itself — audio in, speech understood, LLM reasoning, tools called, speech out — and everything around it: the telephony that gets the call connected, the dashboard a client uses to configure their agent, and the webhooks and CRM plumbing that turn a call into a lead.
This is not a feature-ticket role. It's a build-and-own role. You'll own the voice stack end to end, from the socket that streams audio to the screen a client uses to change their agent's greeting. When a call stutters, drops, or talks over the customer, you're the one who finds out why — and fixes it.
What you'll do
Voice pipeline & real-time systems
- Build and harden the real-time calling pipeline: telephony ↔ audio streaming ↔ speech/real-time models ↔ tool calls ↔ speech out, with latency you can't feel.
- Own the hard voice problems: turn-taking and barge-in, silence and voicemail detection, noise and echo, language switching mid-call (English ↔ Hinglish ↔ regional), and graceful recovery when a model or carrier hiccups.
- Integrate and evaluate voice models and providers across STT/TTS, real-time LLMs, VAD, and telephony carriers — and help decide what we build on next.
- Implement non-blocking tool calling inside live calls — lookups, bookings, CRM writes — so the agent never goes silent or loses the thread.
Platform & full-stack product
- Build the product around the calls: agent configuration (tasks, behaviour, guardrails, AI config), call logs and transcripts, analytics, and bulk-calling campaign controls.
- Design and ship the Node.js/Express services and APIs — call orchestration, queueing, webhooks, retries, recording and transcript storage — that hold up as call volume grows.
- Wire calls into the rest of the platform: CRM contacts and tickets, WhatsApp follow-ups after a call, calendar and booking integrations, Google Sheets and Meta lead-form ingestion.
- Ship clean, usable Angular frontends for our team and for clients — the agent configurator, the inbox, the call review screens.
Reliability, cost & scale
- Instrument everything: time-to-first-word, per-call latency breakdowns, drop rates, tool-call failures, model cost per minute. Build the dashboards that tell us something's wrong before a client does.
- Bring down cost per minute without hurting quality — model choice, streaming strategy, caching, carrier routing.
- Debug production incidents on real client calls, reproduce them, and fix them at the root.
Working with the team
- Pair with the prompt engineer and customer success on what the platform needs for agents to behave well — new tool types, better fallbacks, a proper test harness for prompts.
- Turn what you learn from client deployments into reusable platform capability, so the next voice agent takes hours to stand up, not days.
Roughly 80% of your time is hands-on engineering: writing code across the stack, debugging live calls, shipping. The other 20% is with the team and occasionally clients — understanding what's breaking in the field and deciding what the platform should do about it.
We're under 15 people building AI calling agents. This is not a comfortable corporate job. You'll be in the trenches, you'll work hard, and you'll compress ten years of ordinary career growth into two (this means salary growth also, not just more work).
What we're looking for
Two things are non-negotiable. Please apply only if you meet both — they're the first thing we screen for.
- You've shipped real-time voice AI to production. Not a demo, not a hackathon — a system where real people spoke to an AI over a phone line or live audio stream, and you owned the pipeline: streaming audio, speech or real-time models, latency, interruptions, the lot. Experience at a Voice AI company is a strong signal.
- You're a full-stack engineer who can build the whole thing. Backend services, APIs, real-time transport, databases, and a frontend a client can use without a walkthrough. You don't need a separate team to ship a feature end to end.
Beyond that, we don't care much about degrees or years. We care that you can:
- Work fluently across the MEAN stack — MongoDB, Express, Angular, Node.js — with TypeScript throughout.
- Build and debug real-time systems: WebSockets, WebRTC, streaming audio, event-driven architectures, queues.
- Integrate with telephony/SIP providers and messaging APIs (WhatsApp Business API) and handle their quirks in production.
- Work with LLM APIs, tool/function calling, and speech models, and reason about latency, cost, and failure modes — not just accuracy.
- Design MongoDB data models and APIs for a multi-tenant SaaS: contacts, conversations, tickets, workflows, role-based access.
- Measure and profile before you optimise, write code others can read, and document what you built.
Nice to have: experience with real-time voice frameworks or APIs (OpenAI Realtime, LiveKit, Pipecat, or similar); Indian telephony providers; Hinglish or Indic speech models; GCP/AWS at scale.
Full-time, on-site in Noida.
Why JoyzAI?
✨ Utmost freedom and autonomy — to make decisions. You are the taskmaster, do it the way you think is best. We will only suggest.
✨ Complete responsibility. You are the taskmaster. Take responsibility for it.
✨ Top-of-the-market pay – we reward high performance with the best packages.
✨ 5-day working – weekends are yours to recharge.
✨ Work from office in Noida – collaborate, learn, and grow with a passionate team.
✨ High-growth startup – opportunity to see first hand how AI brings change and be a part of that.