Cartesia
Software Engineer, Agent Harness
*HQ - San Francisco, CA. Posted 6 months ago
The role
You will build the platform behind Managed Agents. You will own the stack end-to-end to shape latency, reliability, and developer experience for real-time voice agents.
What you will do
- Evolve the realtime transport layer for audio between users and agents.
- Trace and cut end-to-end conversational latency across the STT, LLM, and TTS pipeline.
- Make agent deployments fast and resilient.
- Define how voice agent quality is measured by building eval suites, LLM-as-judge scoring, and regression tracking.
Required qualifications
- Experience building and operating distributed backend systems in production.
- Strong product instincts regarding customer-facing latency, reliability, and developer experience.
- Fluent use of coding agents.
- Clear communication with technical and non-technical audiences.
- Ability to write sharp specs and maintain a presence in design reviews.
Nice to have
- Experience with real-time media systems in production, including WebRTC, SIP, RTP, or low-latency audio.
- Experience with voice agents or agent frameworks such as Pipecat, LiveKit, OpenAI Agents SDK, Claude SDK, or Vercel AI SDK.
- Experience building developer-facing platforms or infrastructure, such as SDK backends or API platforms.
- Experience with distributed systems at scale, including high-throughput workloads, autoscaling, multi-region architectures, or queue-based systems.
Benefits
- Competitive base salary
- Attractive equity package
- Fully covered medical insurance along with dental and vision for you and your family
- 9 weeks paternity leave
- 12 weeks maternity leave
- 401(k)
- Monthly commuter allowance
- Flexible PTO
- Lunch, dinner and plenty of snacks provided daily
- Your own personal Yoshi
Before you apply
- Visa sponsorship is provided and assessed on a case-by-case basis.
- Cartesia participates in E-Verify and will provide Form I-9 information to confirm employment eligibility after hire.
About the company
Cartesia builds real-time multimodal intelligence through voice models and a production platform. The founding team met as PhDs at the Stanford AI Lab where they invented State Space Models.