Senior AI Voice Engineer

We are looking for a Senior AI Voice Engineer with strong hands-on experience building real-time conversational voice systems and telephony infrastructure. You will be responsible for developing a standalone AI voice-agent service that integrates with an existing SaaS platform and can autonomously make and manage real-world phone calls. This is not a simple chatbot, LLM integration, or prompt-engineering project. The focus is on building a reliable, low-latency voice system capable of handling interruptions, IVRs, DTMF, voicemail, transfers, silence, external API calls, and other unpredictable telephony conditions. Technology Stack The expected technology stack includes: Python Pipecat FastAPI Twilio Voice / Twilio Media Streams OpenAI Realtime API and/or OpenAI models Speech-to-text and text-to-speech providers WebSockets / REST APIs Docker Async Python We are particularly interested in engineers who understand real-time audio pipelines and telephone systems, rather than developers whose experience is limited to integrating LLMs into conventional web applications. What You Will Build You will own the development of a standalone voice-agent service that communicates with our existing application through a well-defined API. The system will need to: Initiate and manage outbound AI-powered phone calls through Twilio Maintain low-latency, real-time voice conversations Handle natural turn-taking and interruptions Detect and respond to barge-ins Navigate IVR systems and send DTMF tones Handle ringing, silence, hold music, voicemail, transfers, and unexpected disconnects Invoke external application functions while a call is in progress Collect and validate structured information from conversations Support configurable prompts, voices, languages, and agent behaviors Generate transcripts and structured call outcomes Report call metrics, events, errors, and other operational data Provide a reusable architecture for future voice-agent workflows The system must be designed for production reliability, not simply to demonstrate a successful proof of concept. What We Provide For shortlisted candidates, we will provide: Detailed technical specifications Functional requirements and acceptance criteria API documentation Mock APIs and development servers Synthetic test data Agent prompts and expected outputs Required third-party development credentials Engineering support for integration-related questions Development will take place in a standalone repository. Access to our production application or customer data will not be required. Expected Deliverables The project will include: Production-ready Python voice-agent service Modular and configurable agent architecture Twilio telephony integration Real-time audio pipeline Dockerized development and deployment environment Automated unit and integration tests End-to-end voice-call testing Error handling, retries, logging, and monitoring Technical and deployment documentation Final engineering handoff The work will be organized into clearly defined milestones. Ideal Candidate You should have substantial experience with several of the following: Pipecat or similar real-time voice frameworks Twilio Voice and Media Streams WebSocket-based audio streaming OpenAI Realtime API Deepgram, Cartesia, ElevenLabs, or comparable STT/TTS services Voice activity detection Barge-in and interruption handling IVR and DTMF automation Async Python FastAPI Function/tool calling Real-time distributed systems Production observability and monitoring Retry and failure-recovery strategies Docker and cloud deployment Hands-on experience building production telephone agents is highly preferred. General LLM, chatbot, RAG, or prompt-engineering experience alone is not sufficient for this role. How to Apply Please do not submit a generic or AI-generated proposal. Instead, describe the most technically challenging real-time voice or telephony system you have personally built. Please include: What the system did Your specific responsibilities The technologies and services you used How you handled real-time audio and interruptions How you dealt with telephony failures or unusual call states Any latency, scalability, or reliability challenges you encountered If your experience is primarily with managed platforms such as Vapi, Retell, Bland, Synthflow, or similar services, please be transparent about that. Such experience is valuable, but this project requires building and controlling a significant portion of the underlying voice infrastructure directly. Shortlisted candidates will receive the complete technical specification before the final scope, milestones, and pricing are agreed upon.