NLP Data Scientist

Role overview

Responsibilities

• Lead fine-tuning with PEFT methods, including LoRA, QLoRA, Prefix Tuning, and Prompt Tuning, for open-source architectures such as Llama and Mistral.
• Prepare high-quality datasets through deduplication, tokenization, synthetic data generation, and human-in-the-loop validation.
• Implement RLHF or DPO alignment for safety, helpfulness, and tone.
• Use quantization methods such as GGUF, AWQ, and GPTQ to reduce model footprint and memory overhead.
• Build evaluation benchmarks and metrics to audit hallucinations, factual accuracy, and domain alignment.
• Manage distributed multi-GPU training, tensor parallelism, and gradient checkpointing.
• Prepare model checkpoints with MLOps teams for cloud deployment and real-time inference.

Requirements

• 4–8 years of data science or advanced machine-learning engineering experience, including 3+ dedicated years training, evaluating, and fine-tuning NLP systems.
• Expertise in Python, PyTorch, transformer architectures, Hugging Face Transformers, Accelerate, and PEFT.
• Understanding of attention, tokenization constraints, context-window degradation, loss optimization, and compute limitations including CUDA.
• Mandatory credential: Google Cloud Professional Machine Learning Engineer or AWS Certified Machine Learning – Specialty.

Preferred qualifications

• A Master’s or Ph.D. in computer science, data science, computational linguistics, or an adjacent quantitative field, with research focused on neural-network text models, is preferred.
• Experience with custom embedding models or domain-specific classification layers in constrained enterprise runtimes is preferred.

Workplace

• The posting explicitly describes the role as remote and lists the work location as offshore.

Employment