Description
Summary We are looking for an experienced AI voice engineer or team to help us migrate from our current Retell AI/Twilio-based voice agent to a fully custom voice agent infrastructure. Project Goal Our primary objective is to significantly reduce the per-minute operational cost of our current voice agent. While Retell AI has been a good solution, the ongoing usage costs are too high for our expected call volume. We want to build a scalable, production-ready inbound AI voice agent using custom infrastructure that gives us greater control over performance, integrations, and operating costs. Proposed Technology Stack We are currently considering using: LiveKit (real-time audio streaming) Deepgram (speech-to-text) Custom LLM orchestration Custom text-to-speech solution (open to recommendations) SIP/telephony integration Other components as required We are open to better architectural suggestions if you believe there is a more efficient or cost-effective approach. Use Case The AI agent will handle inbound customer calls for appointment booking and customer service. The system should be capable of: Answering inbound calls Understanding natural conversations Booking appointments Handling common customer inquiries Managing conversation context Escalating or transferring calls when required In future phases, we also plan to integrate multiple third-party platforms such as CRM systems, scheduling platforms, and other business software so the AI agent can create, modify, and manage appointments automatically. What We're Looking For We're looking for someone with proven experience building production-grade conversational AI voice systems—not someone who has only configured Retell or Vapi. Experience with technologies such as: LiveKit Deepgram OpenAI or other LLMs SIP/Twilio telephony Voice activity detection Streaming pipelines Low-latency voice systems Function calling/tool integrations Scalable backend architecture is highly preferred. Proposal Requirements Please include the following in your proposal: Your recommended architecture for this project. Whether you would use our proposed stack or recommend a better alternative (and why). Expected latency and performance. Estimated infrastructure costs compared to Retell AI. Estimated development timeline. Your previous experience building similar AI voice agents. Links to relevant projects, demos, GitHub repositories, or case studies. Any suggestions for improving scalability, reliability, or reducing long-term costs. We are looking for someone who can think beyond implementation and help us build a cost-efficient, scalable voice AI platform for long-term use. If you've successfully built custom AI voice agents at scale, we'd love to hear from you.