Staff Software Engineer, Voice Agent

Decagon Decagon · Vertical AI · San Francisco, CA · Engineering

Staff Software Engineer, Voice Agent at Decagon, focusing on the architecture and evolution of the real-time voice platform for conversational AI customer experiences. The role involves leading projects to improve speech understanding, synthesis quality, timing, responsiveness, and reliability across millions of interactions, collaborating with Research, Infra, and Product teams. Responsibilities include defining standards for reliability, testing, and observability, building debugging frameworks, and mentoring senior engineers.

What you'd actually do

  1. Own the architecture of Decagon’s real-time voice runtime and shape its long-term roadmap
  2. Lead initiatives that improve speech understanding, synthesis quality, and conversational timing
  3. Define reliability, testing, and observability standards for live voice interactions
  4. Build frameworks that make voice systems easy to debug, measure, and iterate on
  5. Mentor senior engineers and help expand the technical culture of the Voice group

Skills

Required

  • 8+ years of engineering experience with significant technical leadership
  • Expertise in real-time systems, streaming pipelines, or audio-based applications
  • Ability to define architectural direction and lead cross-functional projects
  • Strong debugging skills across audio, networking, and model-driven systems
  • Experience mentoring engineers and influencing engineering standards

Nice to have

  • Experience with speech recognition or synthesis systems
  • Experience with VAD, streaming protocols, or other real-time audio systems
  • Experience designing or maintaining LLM-driven applications
  • Experience optimizing performance for low-latency use cases

What the JD emphasized

  • real-time systems
  • speech understanding
  • synthesis quality
  • conversational timing
  • reliability
  • testing
  • observability
  • low-latency

Other signals

  • real-time systems
  • speech understanding
  • synthesis quality
  • conversational timing
  • LLM-driven applications