In this role, you will build software to support the machine learning development cycle from data generation, to training models, to deployment and monitoring production machine learning systems in real customer environments.
At Pangram, ML engineers are highly involved in the research effort, are involved in publishing research, and regularly contribute ideas and innovations to the team. However, formal research experience is not necessary. This is an in-person role in our office in Downtown Brooklyn, NYC.
Responsibilities:
- Build robust data pipelines that mine the Internet at scale and generate millions of synthetic text examples for training detection models
- Manage distributed infrastructure for multi-GPU LLM training
- Profiling and optimizing training and inference code
- Deploy efficient inference pipelines for serving LLMs at scale
Requirements:
- B.S. or M.S. in Computer Science or related areas
- Practical experience with deep learning: internships, undergrad or masters' level research projects in an academic lab, Kaggle competitions, or interesting side projects
- Strong programming skills in Python and modern ML frameworks
- Excellent understanding of transformers and LLM fundamentals
- Comfort working across research and engineering boundaries
Nice to have
- Experience with NVIDIA GPU programming and CUDA
- Experience with distributed training frameworks, such as DeepSpeed, FSDL, Ray
- Experience with inference frameworks like vLLM
- Experience with large-scale data processing (Spark, Beam) and orchestration (Airflow)
- Experience with MLOps and experiment tracking
- Experience with DevOps tools
- Familiarity with cloud-based infrastructure (AWS/GCP)
