Top AI & Machine Learning Jobs in San Francisco, CA

Design and implement scalable, secure inference services for speech processing. Optimize networking, audio transcoding, latency, memory, and distributed compute orchestration. Debug complex system issues, customize backend services for customers, and partner with Product to deliver end-to-end features.

Own technical and strategic partner relationships for infrastructure platforms (silicon, cloud, edge, confidential computing). Lead onboarding, POCs, demos, deployment guidance, troubleshooting, and joint go-to-market to drive adoption and expansion of Deepgram’s voice AI on partner hardware and platforms.

Own the research-to-production pipeline: harden training/evaluation workflows, build packaging/deployment tooling, design release gates, optimize inference and serving for latency/throughput, establish benchmarking and validation, and instrument feedback from production to accelerate research iteration.

Analyze and characterize conversational audio data, identify underrepresented areas, and build active-learning and human-in-the-loop workflows. Develop representative benchmarks, automate model adaptation, and create reusable data analysis and scoring systems. Partner with research, engineering, and data teams to run experiments, improve models, organize data, and translate complex findings into measurable product improvements.

Take Deepgram's speech models and optimize, compile, and run them on resource-constrained embedded and edge hardware. Work across model compression, runtime performance (C/C++/Rust), RTOS/bare-metal and embedded Linux environments, vendor NPUs/DSPs and inference runtimes, benchmarking, deployment pipelines, OTA updates, and partner with silicon vendors to enable real-time, low-power on-device voice experiences.

As an ML Ops Infrastructure Engineer, you will design CI/CD pipelines for ML, maintain deployment systems, and implement monitoring while collaborating with research teams to ensure model quality and performance.

Own the end-to-end product experience for AI coding agents integrating Deepgram: discovery, integration/onboarding, production use, verification, and feedback systems. Build measurement, experimentation, and optimization systems; define agent-specific developer platform requirements and prototype changes; turn agent traffic into rapid product improvements and run cross-functional execution.

Design and build scalable, automated evaluation pipelines and infrastructure to validate speech, audio, and multimodal models. Translate research benchmarks into enforceable pass/fail gates, run batch and streaming tests, operate canaries and continuous monitoring, integrate quality gates into CI/CD, and partner with research, training, inference, and infra teams to prevent regressions and ensure production model quality.

Lead end-to-end design and execution of voice data programs: build ingestion pipelines, labeling specs, tooling, and quality safeguards; align data strategy with research and product; manage vendors and teams; track throughput, quality, and vendor performance; and prototype/deploy data tools to scale datasets that improve speech and voice AI model outcomes.