Research Scientist Intern, FM Quality ML

About the role#

The FM Quality ML team works at the intersection of machine learning and system optimization within the Facebook Ads Ecosystem. This team focuses on how to deliver ads that balance revenue growth with user experience across Feed, Videos, Reels, and Stories. As an intern, you will frame practical ads quality problems as machine learning challenges and build solutions alongside research scientists and software engineers. The core task involves building a quality system that understands user history and ad content to predict quality signals while maintaining strict latency and compute constraints.

What you'll do#

• Post-train large language models using supervised fine-tuning and reinforcement learning to predict ad quality signals based on user behavior and ad content.
• Design reasoning formats, model inputs, and reward functions to improve the accuracy and calibration of quality predictions.
• Create evaluation methods for LLM reasoning, such as human-anchored benchmarks, LLM-as-a-judge, groundedness checks, and comparisons against production models.
• Distill model reasoning into efficient production systems and validate results through offline metrics and online experiments.
• Communicate research plans and results while collaborating with cross-functional partners.
• Publish research findings and contribute to work that supports product development.

What you'll need#

• Currently pursuing a Ph.D. in Computer Science, Artificial Intelligence, Machine Learning, Natural Language Processing, Information Retrieval, or a related field.
• Proficiency in Python, C++, C, or Java.
• Experience with deep learning frameworks like PyTorch or TensorFlow.
• Familiarity with generative model training and deep learning architectures such as Transformers, GPTs, or BERT.
• Experience building systems using machine learning, information retrieval, or natural language processing.
• Ability to maintain work authorization in the country of employment.
• Intent to return to your degree program after the internship.
• A record of research achievements through publications, patents, grants, or fellowships.
• Prior experience in areas like post-training of large language models, reward design, uncertainty estimation, knowledge distillation, or personalized recommendation systems.
• Demonstrated software development experience via internships, work experience, or open source contributions.

Location & details#

• Location: Menlo Park, California.
• Term: Rolling start dates with a duration of 12, 16, or 24 weeks.
• Work Modality: On-site.
• Employment: Full-time internship.