Experienced Machine Learning Engineer (Speech Recognition, LLM Fine-tuning) – Remote

About the role

Noble Cortex is seeking a Machine Learning Engineer with a passion for speech recognition and Large Language Model (LLM) fine-tuning. This role provides the chance to work on cutting-edge AI projects that push the boundaries of communication technologies and unlock the potential of AI-driven solutions.

What You’ll Do

• Develop Speech Recognition Models: Design, build, and fine-tune speech recognition models using the latest advancements in deep learning and AI.
• Fine-tune LLMs: Leverage state-of-the-art techniques like LoRA to fine-tune large language models to domain-specific tasks and optimize model performance.
• LLM Scaling & Optimization: Tackle challenges in model scaling, memory efficiency, and performance optimization for real-time applications.
• Domain Adaptation: Use transfer learning techniques to adapt general LLMs to specific domains or tasks, such as fine-tuning for specific dialogue systems, NLP tasks, or industry-specific applications.
• Collaborative Problem-Solving: Work with cross-functional teams to integrate AI models into production environments and ensure real-world effectiveness.

Key Responsibilities

• Deep knowledge of transformer-based architectures (e.g., GPT, BERT, T5, LLaMA), with a focus on model pre-training, fine-tuning, and optimization.
• Hands-on experience with LLM fine-tuning, LoRA, parameter-efficient tuning, and techniques for improving generalization and task adaptation.
• Experience in speech recognition systems, including ASR models and audio processing techniques.
• Strong familiarity with real-time optimization techniques and the challenges of deploying LLMs in low-latency environments.
• Comfort with experiment tracking and hyperparameter tuning for training large models at scale.

Additional Experience We’d Like

• Working with LLMs beyond traditional fine-tuning, such as RLHF or using techniques to align models with human preferences.
• PRACTICAL experience with distributed training and fine-tuning frameworks like Hugging Face Transformers, DeepSpeed, Accelerate, or Megatron-LM.
• Knowledge of data-efficient training methods, including few-shot learning, zero-shot learning, and active learning.