LLM Scientist

We are looking for an "LLM Scientist" for our client, an international AI development company based in New York, who will collaborate closely with Data Engineers, enhancing the quality of data and the efficiency of data pipelines, and contribute to a team culture that values mentorship and knowledge sharing.

Key Responsibilities Design and execute experiments by conducting ablation studies on various prompt strategies, tokenization methods, and fine-tuning hyperparameters. Own model decisions regarding the selection of foundation models for specific tasks and the acceptable trade-offs between latency and accuracy. Architect data pipelines in collaboration with Data Engineers to curate high-quality synthetic data and improve model training efficiency. Solve complex 'hallucination' problems by serving as the final escalation point for enhancing model factuality and designing logic for retrievers and re-rankers. Mentor the team by translating the latest research papers into actionable engineering tickets.

Qualification Required Extensive experience with PyTorch and the Hugging Face ecosystem (Transformers, PEFT, Accelerate). Proven track record of fine-tuning models, with a deep understanding of attention mechanisms, positional embeddings, and transformer architectures. Hands-on experience with vector databases (Pinecone, Milvus) and familiarity with the mathematics behind similarity search and embedding models. Proficiency in Python, along with familiarity in containerization (Docker) and cloud-scale Al training (AWS SageMaker/GCP Vertex Al). Ability to analyze model failures, hypothesize root causes, and systematically test solutions.

Nice to Have Advanced Degree: A Master's or PhD in Computer Science, Math, or a related field with a focus on NLP. Multi-Modal Experience: Familiarity with Vision-Language Models (VLM) or audio-to-text integration. Reinforcement Learning: Experience with RLHF (Reinforcement Learning from Human Feedback) or DPO (Direct Preference Optimization). Open Source Presence: A portfolio of public GitHub repos, Hugging Face models, or published papers in the GenAl space. Hardware Knowledge: Deep understanding of NVIDIA GPU architectures (H100/A100) and CUDA kernels to maximize hardware utilization.

About Our Client Our client is a people-focused organization dedicated to developing impactful products and services that create meaningful value for customers and communities. They foster a collaborative, respectful, and inclusive work environment where employees are encouraged to take ownership, contribute ideas, and grow professionally. The company supports flexibility and work–life balance while maintaining strong performance and accountability standards.

Benefits & Wellbeing Compensation for this role is determined based on competitive market data and may vary depending on geographic location, experience, skills, and qualifications. Specific details will be discussed during the interview process. The company offers a comprehensive benefits package designed to support employees’ well-being, financial security, and professional development. Benefits may include medical, dental, and vision coverage, retirement plan contributions, paid time off, flexible working arrangements, and opportunities for career growth, in accordance with company policies and applicable local regulations. It is an equal opportunity employer and considers all qualified applicants without regard to legally protected characteristics. Applicants must have the legal right to work in the country of employment.