About

We’re looking for an expert in advanced prompt engineering who can design and optimize voice-first AI agents. The core of this project is not just wiring together APIs - it’s about building prompting architectures that make a voice agent: Factually accurate (knowledge base–driven). Natural and conversational (low latency, human-like dialogue). Reliable in real-world scenarios (minimizing hallucinations, handling edge cases). The agent will run on modern voice orchestration platforms (Daily.co, Synthflow, VoiceGenie, Retell, etc.), but the true differentiator will be your ability to engineer prompts, guardrails, and multi-step orchestration flows that ensure consistent, production-level performance. Core Responsibilities: Architect system prompts and conversation flows for voice-based LLMs. Develop context injection and RAG-based prompting strategies tied to our knowledge base. Design multi-turn conversation logic with fallback flows and escalation triggers. Implement function/tool calling prompts (API lookups, ticket creation, routing). Optimize prompts for low latency voice pipelines (STT → LLM → TTS). Continuously refine through red-teaming, A/B testing, and hallucination minimization. Must-Have Skills: Deep expertise in prompt engineering: prompt chaining, guardrails, system vs user prompts, structured outputs. Experience designing for voice-first interactions (not just text chat). Familiarity with voice orchestration tools (Daily.co, Synthflow, Retell, VoiceGenie). STT/TTS model experience: Deepgram, Whisper, ElevenLabs, Play.ht. Knowledge base integration (vector databases like Pinecone, Weaviate, or Neo4j). Ability to balance latency vs accuracy in voice conversations. Backend experience (Node.js/Python) for implementing orchestration logic. Nice-to-Have: Telecom/VoIP knowledge (SIP, WebRTC, Twilio/Vonage integrations). Production deployments of voice AI agents at scale. Experience with LangChain, LangGraph, or custom orchestration frameworks. Engagement Type: Project-based. Potential for ongoing work as we scale multiple agents. What Success Looks Like: A production-ready AI voice agent that: Responds accurately using our knowledge base. Handles multi-turn conversations smoothly. Executes functions reliably via tool calling. Uses prompting best practices to minimize hallucinations. Delivers a human-like, low-latency voice experience.