Job Description

Responsibilities

• Assess client-specific data assets and determine the appropriate adaptation strategy — continued pretraining, supervised fine-tuning, or a combination — based on the domain, data volume, and use case requirements

• Curate, clean, structure, and prepare domain-specific datasets from raw client data for use in model training pipelines

• Fine-tune large language models in the 70B–100B+ parameter range using techniques such as LoRA, QLoRA, and multi-adapter patterns

• Perform continued pretraining on open-weight models (Qwen, Llama, and related ecosystems) to embed domain knowledge directly into model weights

• Manage distributed training workflows across multi-node GPU clusters

• Design and execute evaluation frameworks to validate domain adaptation quality, factual grounding, and model behavior

• Support RAG system development where applicable, including vector database integration, chunking strategies, and reranking pipelines

• Contribute to inference optimization and deployment pipeline integration

Required Qualifications

• Currently enrolled in or recently completed a Bachelor's, Master's, or PhD program in Computer Science, Machine Learning, or a related field

• Demonstrated hands-on experience fine-tuning large language models, supported by concrete project work, research, or open-source contributions

• Experience with frontier-scale models (100B+ parameters) or distributed training across multi-node GPU clusters

• Familiarity with parameter-efficient fine-tuning methods (LoRA, QLoRA) and open-weight model architectures

• Experience with data curation and preparation workflows for LLM training, including cleaning, formatting, deduplication, and quality filtering

• Proficiency in Python-based ML frameworks such as PyTorch, HuggingFace Transformers, DeepSpeed, or FSDP

• Understanding of training compute, memory constraints, and inference trade-offs at scale

Nice to Have

• Familiarity with RAG architectures or production inference serving frameworks (vLLM, TGI, TensorRT-LLM)

• Experience in low-resource or multilingual NLP settings
• Relevant publications, open-source contributions, or documented projects involving LLM training