Remote AI Research Architect (Multimodal LLMs)

Tether Operations; Poland; Remote; PLN 240,000 - 360,000; Full time

Benefits offered by this job

Remote work

Job summary

Tether is seeking a researcher to join the AI model team and drive innovation in architecture development for large language models and multimodal systems. You will work on pre-training, alignment, and data pipelines at scale, with a hands-on, research-driven approach and a strong publication record.

Requirements include PhD in NLP/ML or related field, deep expertise in PyTorch and HuggingFace, and experience with thousands of GPUs in distributed training environments.

Qualifications

• PhD in NLP/ML with AI R&D publications is preferred.
• Experience with large-scale LLM or multimodal pre-training on thousands of GPUs.
• Familiarity with distributed training frameworks and libraries.
• Deep knowledge of transformer architectures and optimization.
• Strong PyTorch and HuggingFace expertise; model development and deployment.

Responsibilities

• Large-Scale Pre-Training: conduct foundational pre-training for LLMs and Multi-Modal models on distributed GPU clusters.
• Architecture & Alignment Innovation: design, prototype, scale architectures, tokenizers, and cross-modal alignment layers.
• Data Strategy: source and curate massive-scale textual and multimodal datasets, establishing robust data pipelines for pre-training.
• Experimental Research: run experiments, analyze results, and refine training methodologies.
• Optimization & Debugging: identify bottlenecks in performance and stability during long training runs.
• System Scalability: contribute to distributed training systems to ensure scalability and efficiency.

Skills

PyTorch

Hugging Face

LLM / Multi-Modal

Distributed Training

Research & Publications

Education

PhD in NLP/ML

CS degree

Tools

CUDA

GPU clusters

Job description

Tether is seeking a researcher to join the AI model team and drive innovation in architecture development for large language models and multimodal systems. You will work on pre-training, alignment, and data pipelines at scale, with a hands-on, research-driven approach and a strong publication record.

Requirements include PhD in NLP/ML or related field, deep expertise in PyTorch and HuggingFace, and experience with thousands of GPUs in distributed training environments.