About The Team
How can we learn meaningful representations of users from their behavior over time and their connections with others?
Our Frontier Model Research team explores this question through behavioral sequence modeling, graph learning, and self-supervised learning. We develop reusable user representations and predictive models that support credit risk assessment, fraud detection, and other financial decision-making applications.
As an intern, you will work closely with researchers and machine learning engineers to implement models, reproduce research ideas, conduct experiments, and help bring validated methods into practical use. Projects may focus on sequence modeling, graph learning, or their downstream applications, depending on your background and the team’s priorities. You are not expected to specialize in both areas before joining.
What You Will Gain
- Mentorship from engineers working on behavioral sequence models, graph learning, model applications, and research diagnostics.
- Practical experience connecting research ideas with real-world data, rigorous experiments, and model integration.
- The opportunity to own a scoped project, explain both successful and unsuccessful experiments, and present your findings to the team.
Job Description
- Develop sequence and graph learning models: Implement and experiment with approaches such as Transformer-based behavioral sequence models, graph neural networks, and self-supervised learning. Investigate input representations, training objectives, model architectures, and sampling strategies within a well-defined project.
- Prepare data for model development: Help construct behavioral sequences, relationship graphs, training samples, and node or edge features. Investigate data quality, temporal consistency, and how data preparation choices affect model performance.
- Reproduce research and run systematic experiments: Read selected papers, implement relevant methods, establish baselines, and conduct ablation studies. Maintain clear records of code, configurations, datasets, and experimental results.
- Analyze and diagnose models: Investigate training behavior, learned representations, and failure cases. Help identify issues related to convergence, information leakage, generalization, or the contribution of different data sources and model components.
- Validate downstream applications: Generate user embeddings or model scores, integrate them into downstream prediction tasks, and evaluate their incremental value using appropriate baselines, temporal validation, and segment analysis.
- Support inference and deployment workflows: Work with application and engineering colleagues on reproducible training and evaluation pipelines, batch inference, model integration, testing, and monitoring. Help troubleshoot issues and improve the reliability and efficiency of model workflows.
Requirements
- Currently pursuing a Bachelor’s or Master’s degree in Computer Science, Artificial Intelligence, Mathematics, Statistics, Data Science, Engineering, or a related field.
- Solid foundations in machine learning and basic deep learning concepts, including supervised learning, optimization, overfitting, and model evaluation.
- Good Python programming skills and hands-on experience with PyTorch or a comparable deep learning framework.
- At least one machine learning or deep learning project from coursework, research, an internship, or independent work. You should be able to explain your personal contribution, modeling choices, experiments, and findings.
- Familiarity with data preparation and analysis using tools such as SQL, pandas, or NumPy.
- Interest in at least one of the following: sequence modeling, graph learning, representation learning, self-supervised learning, or recommendation systems.
- Ability to read technical materials and research papers in English, communicate findings clearly, and collaborate with others.
- A careful and curious approach to experimentation, with a willingness to investigate unexpected results, debug code, and document findings.
- Project or research experience with Transformers, user behavior modeling, graph neural networks, heterogeneous graphs, link prediction, or self-supervised learning is preferred.
- Experience with graph learning frameworks such as PyTorch Geometric or DGL is preferred.
- Exposure to large-scale data processing, GPU training, reproducible ML pipelines, batch inference, or model deployment is preferred.
- Research experience or open-source contributions demonstrating practical implementation and experimental analysis is preferred.