Senior Applied AI Scientist - NLP

About the Role

CyberTide GmbH is seeking an experienced Applied AI Scientist to join our team. This role focuses on designing, fine-tuning, and optimizing transformer architectures, particularly open-source LLM models, with a strong emphasis on efficiency, robustness, and performance optimization.

What You Will Do

• Architect, fine-tune, and optimize existing open-source LLMs (e.g. Mistral, Llama, Gemma, BERT, etc) for specific tasks and efficiency.
• Enhance model performance using advanced techniques such as parameter-efficient fine-tuning (e.g., LoRA, QLoRA) and transformer architecture optimization.
• Focus on NLP-specific challenges, including language understanding, generation, classification, sequence-to-sequence tasks, and developing novel architectures or improving existing ones to enhance NLP model performance.
• Research, experiment, and innovate in model architecture optimization, focusing on enhancing transformer-based models for better scalability and efficiency.
• Develop and implement efficient training pipelines for transformer models, including data preprocessing, augmentation, distributed training, and mixed-precision training.
• Apply advanced prompt engineering techniques to improve model outputs and alignment with user requirements.
• Collaborate with clients to deliver tailored technical solutions, translating customer requirements into effective AI approaches.
• Ensure AI models are designed to enhance user experience and address end-user needs effectively.
• Collaborate closely within a small, agile startup team, where your ideas and expertise will have a direct impact. Exhibit strong communication skills to articulate concepts clearly and contribute meaningfully across diverse functions.

Your Profile

• At least 3 years of experience in machine learning, Deep learning and data Science, with a minimum of 1 year working specifically with LLMs, fine-tuning, transformer architecture optimization.
• Proven experience in fine-tuning and optimizing transformer-based models, including but not limited to Mistral, LLaMA, Gemma, BERT, and other open-source architectures.
• Proficiency in Python and deep learning frameworks such as PyTorch or TensorFlow.
• Familiarity with the Hugging Face ecosystem, including Transformers library and Model Hub. Contributions to the Hugging Face community or experience with Hugging Face Spaces is a plus.
• Experience with training efficiency optimization techniques, including mixed-precision training, model pruning, quantization, and knowledge distillation.
• Knowledge of AI model inference techniques and frameworks, including Retrieval-Augmented Generation (RAG) for enhancing contextual understanding and response generation.
• Fluency in English. German or French language skills are preferred, as they align with our target market.
• Ability to continuously learn and adapt cutting-edge techniques to enhance model performance and deployment efficiency.
• Thrive in a fast-paced startup environment where rapid iteration and efficient delivery are essential to success, ensuring consistently high-quality results within tight deadlines.