Multimodal ML Engineer — Ship Vision, Audio & Video AI Safety

Moonfire; Greater London; Hybrid; GBP 120,000 - 170,000; Full time

Benefits offered by this job

Equity

Flexible time off

Paid time off

Relocation package

Job summary

White Circle is an AI Safety company building the safety, reliability, and optimization layer for AI systems. The team trains and deploys multimodal models across vision, audio, and video modalities to power safety policies at scale.

We seek an experienced Multimodal ML Engineer to lead end-to-end model development, from training from scratch to production serving, and to design robust evaluation benchmarks for real-world AI safety use cases.

Qualifications

• Proficient in PyTorch with experience in distributed training.
• Strong background in multimodal model architectures across vision/audio domains.
• Hands-on experience with RLHF/alignment for multimodal systems.
• Track record shipping models to production with latency targets.

Responsibilities

• Train and fine-tune large-scale multimodal models (vision-language, audio, speech).
• Extend models across modalities including video temporal modeling and streaming audio.
• Design experiments and data mixes; build training recipes.
• Build and maintain multimodal data pipelines and synthetic data generation.
• Train and optimize MoE architectures for efficient inference.
• Deploy models end-to-end from research checkpoint to production serving.
• Define metrics and benchmarks relevant to product needs.

Skills

PyTorch

Distributed training

Multimodal architectures

RLHF/alignment

Video/audio sequence modeling

Production deployment

Engineering fundamentals

Latency optimization

Tools

DeepSpeed

FSDP

Whisper

HuBERT

Job description

White Circle is an AI Safety company building the safety, reliability, and optimization layer for AI systems. The team trains and deploys multimodal models across vision, audio, and video modalities to power safety policies at scale.

We seek an experienced Multimodal ML Engineer to lead end-to-end model development, from training from scratch to production serving, and to design robust evaluation benchmarks for real-world AI safety use cases.