Machine Learning Engineer, Speech LLM Training - San Francisco

Key Responsibilities

• Build and train large-scale audio or speech models (Unified SpeechLLMs, ASR, TTS, Generative Audio).
• Design novel sequence modeling architectures and debug distributed training clusters.
• Traverse the entire stack: from signal processing and raw acoustic representations to foundation model training.
• Optimize large-scale distributed training runs, manage GPU memory, and resolve performance bottlenecks.
• Design and train state-of-the-art neural audio codecs, diffusion models, or flow matching architectures for voice generation.
• Apply RL techniques (RLHF, GRPO) to improve conversational cadence and model steerability.
• Optimize end-to-end inference for real-time cloud streaming using frameworks like vLLM or TensorRT-LLM.
• Manage massive GPU clusters and distributed training frameworks (FSDP, DeepSpeed, Kubernetes).

Requirements

• Proven track record of building and training large-scale audio/speech models from the ground up.
• Deep expertise in PyTorch or JAX with experience in distributed training.
• Ability to work at the intersection of research and engineering.
• Experience with optimization of large-scale GPU memory and performance.
• Thrives in a fast-paced, high-growth startup environment.

Preferred Qualifications

• Experience with Text-based LLMs (pretraining, instruction tuning, RLHF).
• Hands-on experience with Neural Audio Codecs.
• Experience with Generative Architectures (diffusion, flow matching, autoregressive).
• Deep Systems Optimization experience (high-throughput serving frameworks).
• Large-Scale Infrastructure management.

Compensation & Benefits

Salary: $180K - $270K base + performance bonus + Equity.

Health: Comprehensive top-tier healthcare (dental, vision) for employees and dependents.

Retirement: 401(k) with company matching.

Time Off: Unlimited PTO + 13 paid holidays + 12 weeks paid New Parent Leave.

Work Model: Hybrid (Minimum 3x in-office per week in San Francisco).

Perks: Top-of-the-line laptops/workstations, annual offsites, fully stocked office.