Research Scientist Intern, FM Quality ML (PhD)

Meta•Menlo Park, CA

1d•$7,650 - $12,667

About The Position

Meta is seeking a Research Scientist Intern to join the FM Quality ML team within the Facebook Ads Ecosystem. The team applies machine learning, generative AI, reinforcement learning, and large-scale systems to personalize ad quality across Feed, Videos, Reels, and Stories, balancing revenue growth with user experience. Interns will frame practical, high-impact ads-quality challenges as machine-learning problems and develop solutions with guidance from research scientists and software engineers. The research focuses on building a quality system that understands people and ads well enough to reason about long-term user value and help people see ads they like and trust. The team is post-training large language models to reason over people's ad and organic interaction histories together with ad content, producing explanations grounded in that history and calibrated predictions for multiple quality signals. Research challenges include developing post-training methods and reward designs, representing ads and user behavior in a shared semantic vocabulary, evaluating reasoning against people's judgments, and distilling model reasoning into efficient production models subject to latency and compute constraints. Offline evaluations and online experiments are used to assess effects on ad experiences and long-term value. Internships last 12–16 or 24 weeks, with various start dates throughout the year.

Requirements

Currently has or is pursuing a Ph.D. in Computer Science, Artificial Intelligence, Machine Learning, Natural Language Processing, Information Retrieval, or a relevant technical field.
Must obtain work authorization in the country of employment by the time of hire and maintain it throughout employment.
Experience with Python, C++, C, Java, or another related programming language.
Experience with deep-learning frameworks such as PyTorch or TensorFlow.
Experience building systems based on machine learning, deep-learning methods, information retrieval, or natural language processing.
Familiarity with algorithms behind generative-model training and common deep-learning architectures such as Transformers, GPTs, and BERT.

Nice To Haves

Intent to return to the degree program after the internship or co-op.
Research achievements demonstrated through grants, fellowships, patents, or publications at leading conferences such as NeurIPS, ICML, ICLR, ACL, EMNLP, KDD, SIGIR, WSDM, WWW, RecSys, or CIKM.
Prior research or project experience in areas such as LLM post-training, reasoning and chain-of-thought methods, reward design, LLM evaluation, uncertainty estimation and calibration, knowledge distillation, sequence modeling, learned content identifiers, multimodal understanding, personalized recommendation, or long-term user behavior modeling, particularly for recommendation or ads systems.
Software-development experience through technical internships, work experience, coding competitions, or widely used open-source machine-learning contributions.
Experience working and communicating cross-functionally in a fast-paced team, identifying research opportunities in real-world applications, selecting appropriate ML methods, and proactively driving iterations based on analysis of results.

Responsibilities

Post-train large language models using supervised fine-tuning and reinforcement learning to reason over user behavior histories and ad content and predict multiple ad-quality signals.
Design reasoning formats, model inputs, and reward functions to improve the accuracy and calibration of quality predictions.
Develop evaluation methods for LLM reasoning and predictions, including human-anchored rubric-graded benchmarks, LLM-as-a-judge, groundedness checks, and comparisons with production models.
Distill LLM reasoning and predictions into efficient production models and validate gains through offline metrics and online experiments.
Collaborate with researchers and cross-functional partners, communicating research plans, progress, and results.
Publish research results and contribute to research applicable to Meta product development.