Multimodal ML Engineer (AI)

Location: Hybrid work from Paris or London; relocation package available for Paris

Company

hirify.global builds an AI safety platform that tests, enforces, and continuously improves natural-language policies for AI systems.

What you will do

• Train and fine-tune large-scale multimodal models covering vision, language, video, audio, and speech.
• Extend models for image understanding, video temporal modeling, long-context processing, and streaming audio.
• Design experiments involving architectures, data mixes, and training recipes.
• Build multimodal data pipelines, including dataset curation and synthetic data generation.
• Develop alignment pipelines using SFT, DPO, GRPO, and reward modeling across modalities.
• Optimize and deploy models for production through quantization, distillation, batching, streaming, evaluation, and low-latency serving.

Requirements

• 3+ years of experience training large-scale deep learning models in multimodal domains.
• Strong PyTorch skills and hands-on distributed training experience with DeepSpeed, FSDP, or similar tools.
• Deep understanding of multimodal architectures, including vision and audio encoders, projectors, and LLMs.
• Hands-on experience with multimodal RLHF and alignment, including GRPO, DPO, and reward modeling.
• Experience with video or audio sequence modeling, temporal modeling, long-context processing, efficient attention, or streaming inference.
• Track record of shipping production models, plus strong engineering fundamentals in clean code, testing, version control, and documentation.

Nice to have

• Understanding of audio signal processing, including spectrograms, mel features, and noise reduction.
• Experience with MoE architectures and large-scale multimodal dataset curation.

Culture & Benefits

• Competitive compensation package with equity.
• Flexible time off and paid time off aligned with local regulations.
• Hybrid work from Paris or London, with a relocation package for Paris.
• Medical insurance in France, learning and development support, and required hardware, tools, and services.
• Covered subscriptions for AI agents and IDEs, plus team off-sites twice a year.

Hiring process

• 25-minute introductory call with HR.
• Take-home test assignment.
• Technical interview with the Head of Applied Research followed by a 45-minute final conversation with the CEO.