Research Staff, LLMs
DeepgramFull Time
Expert & Leadership (9+ years)
Mountain View, California, United States
Candidates must possess a PhD in Computer Science, Electrical Engineering, or a related field, or an equivalent combination of education, training, and experience, with at least 10 years of research experience in AI/NLP/ML, and experience conducting research and shipping user-facing products. Experience with large language models (LLM) including Transformer model architecture, attention mechanisms, and decoder-only LLMs is required, as is foundational LLM training experience, including data curation, distributed training, and hyperparameter tuning. Experience in making LLM-based solutions deployable on-device with small latency and memory, along with NPU optimization, is also necessary, as is expertise in LLM alignment, instruction tuning, LoRA, and Adapter techniques. Proficiency in deep learning frameworks such as TensorFlow or PyTorch is required, along with hands-on experience with large language models like GPT and LLaMA.
The Senior Staff Research Engineer will conduct cutting-edge research and development of large foundation models (LLM, VLM, and Reasoning) for future Samsung products, including model design, efficient model training, instruction tuning, and prompt engineering. They will collaborate with a multidisciplinary team to understand requirements, develop prototypes, and deliver robust solutions, conduct thorough evaluations of model performance, identify areas for improvement, and propose innovative solutions to enhance large language models. The role also involves generating creative solutions (patents) and publishing research results in top conferences, and contributing to the overall advancement of on-device language intelligence research.
Develops advanced technology in multiple domains
Samsung Research America specializes in advanced technologies such as next-generation communications, artificial intelligence, digital media, mobile platforms, and digital health. Their innovations include FadeNet, a convolutional neural-network based technology, and the ECG Monitor App, which enables ECG recording using Galaxy Watch for atrial fibrillation screening.