Research Scientist Intern, Video RecSys (PhD)

This team is directly in charge of recommending all sorts of videos to users, powered by our large-scale unified video recommendation models. In this role, you will work on Recursive Self-Improvement (RSI): building AI agents that autonomously propose, run, evaluate, and learn from improvements to these models, closing the loop between ML experimentation and model quality. Our interns will have the opportunity to frame practical, high-impact challenges in recommendation into rigorous machine learning problems, and bring their own expertise to build innovative solutions under the guidance of our Research Scientists and Software Engineers.

More concretely, you will develop an agent-driven RSI prototype that continuously proposes, evaluates, and refines improvements for a selected video recommendation task. The scope spans training data improvement, environment and experiment setup, automated offline model evaluation, and feedback collection from model quality signals such as normalized entropy (NE) and recall. You will use experiment outcomes to improve both the agent's intervention strategy and its evaluation criteria, and deliver a working end-to-end prototype with evidence against a fixed baseline. Work will be mentored by the team, with production changes and online experiments subject to owner review and approval.

Our internships are twelve (12) to twenty-four (24) weeks long and we have various start dates throughout the year.

Responsibilities

Design and build an agent-driven Recursive Self-Improvement (RSI) system that continuously proposes, evaluates, and refines improvements to a large-scale video recommendation model. Profile training data quality and signal coverage, and implement bounded training-data interventions such as filtering, sampling, and reweighting. Automate environment and experiment setup, including reproducible training configurations, offline model evaluation pipelines, and evaluation benchmarks. Build feedback loops that collect and attribute model quality signals (e.g., NE and recall gains or losses) and use them to improve the agent's intervention strategy and evaluation criteria. Measure downstream model quality alongside data-processing and training costs, and maintain versioned experiment records and automated checks. Collaborate with researchers and cross-functional partners including communicating research plans, progress, and results. Publish research results and contribute to research that can be applied to Meta product development.

Qualifications

Currently has or is in the process of obtaining a Ph.D. degree in Computer Science, Artificial Intelligence, Machine Learning, or relevant technical field Must obtain work authorization in country of employment at the time of hire and maintain ongoing work authorization during employment Experience with Python, C++, C, Java or other related language Experience with deep learning frameworks such as Pytorch or Tensorflow Familiarity with machine learning and common deep learning architectures Intent to return to the degree program after the completion of the internship/co-op Proven track record of solid research achievements as demonstrated by grants, fellowships, patents, as well as publications at leading AI conferences such as NeurIPS, ICML, ICLR, KDD, RecSys, WWW, ACL and EMNLP Prior research or project experience in one or more of the following areas: LLM-based agents and agentic workflows, recursive self-improvement, automated machine learning (AutoML), data-centric AI (data selection, curation, and reweighting), reinforcement learning, and large-scale recommendation systems Experience building reproducible ML experimentation and evaluation pipelines on large-scale distributed training systems Demonstrated software development experience via tech internships, work experience, coding competitions, or widely used contributions in open source machine-learning repositories Experience working and communicating cross functionally in a fast-paced team environment. Ideal candidates should have the ability to quickly understand and identify the research opportunities behind real-world applications, select the appropriate ML methods to explore, and proactively drive the iterations based on clear analysis of the current results