Software Engineer - Voice AI (Inference Runtime)

ABOUT BASETEN

Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma, and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F, led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to ship AI products.

THE ROLE

Voice is becoming the internet’s next interface, but a production-grade Voice AI system is "hard to build". You’ll join a small founding team of Baseten Voice AI, focused on bringing state-of-the-art open source models into production for Voice AI customers across productivity, customer service, clinical conversation, creator tools, education, and more. You’ll make a meaningful impact on people’s daily lives and help reshape these industries.

This is a high-impact, high-ownership role. You will be the primary owner of Baseten Voice AI - our in-house inference stack to power Voice AI models - from product roadmap through engineering implementation. You’ll partner closely with Forward Deployed Engineers, Model Performance Engineers, and sister engineering teams to push the boundaries of Voice AI.

EXAMPLE INITIATIVES

• Develop world-class model serving stack for state-of-the-art open-source voice models - reduce end-to-end and tail latency (p95/p99), increase throughput, and improve GPU efficiency via profiling, runtime tuning, and server-level optimizations.

• Build large-scale, real-time infrastructure for multi-model voice agents - orchestrate STT, TTS, and agent components with streaming I/O to meet customer SLOs.

• Design tight training and inference iteration loops for voice model customization - enable fast evaluation, safe rollout, and rapid experimentation for custom voice model development.

• Past projects:

  • Baseten leads Coval’s voice AI benchmark

  • How Notion achieved 2-3x lower latency with AI Meeting Notes

  • Cost-efficient, high-performance TTS with Qwen3-TTS

  • Fine-tuning Qwen3-TTS for high-quality voice cloning

  • The fastest Whisper — with streaming and diarization

  • NVIDIA Nemotron 3 Diarization: real-time speaker labels at a cent per audio hour

RESPONSIBILITIES

• Own and lead Voice AI product areas end-to-end - from architecture and system design through implementation, rollout, and long-term production operations.

• Design, build, and operate real-time, large-scale, high-performance model serving systems for STT, TTS, and voice agent workloads for mission-critical customer deployments

• Drive cross-team collaboration with sister engineering teams to solve full-stack technical problems, align on priorities, and coordinate end-to-end delivery across the product surface area

• Mentor teammates through code reviews, design docs, and technical leadership.

REQUIREMENTS

• Bachelor's degree or higher in Computer Science or related field

• Proven track record owning production-grade real-time, large-scale systems where tail latency (p99) matters.

• Proficient coding abilities in one or more popular programming or scripting languages; Python proficiency is a plus.

• Good taste in product, particularly developer-oriented tools

• Interest in ML/AI infrastructure and willingness to learn

• Strong collaboration and communication skills

• Comfortable using AI coding assistants (e.g., Claude Code, Codex, Cursor) as a daily productivity multiplier — as an AI-native company, we see this as a must-have skill.

NICE TO HAVE

• Experience implementing pipeline-level model runtime optimizations such as dynamic batching, async scheduling, or decode-side throughput improvements.

• Experience building developer platforms: SDKs, CLIs, APIs, and self-serve workflows for ML or infrastructure products.

• Experience with containerization and orchestration technologies (Docker, Kubernetes), service meshes, or distributed scheduling.

• Familiarity with speech/audio ML models (STT, TTS, speech-to-speech)

• Familiarity with model-serving runtimes (vLLM, TensorRT, ONNX).

• Familiarity with systems-level performance profiling across host-device boundaries (e.g. PyTorch Profiler), diagnosing GPU utilization issues