ML Research Engineer — Usage, Subscription, and Payment Intelligence

Job description

**About Chargebee** Chargebee is building AI-native billing and monetization infrastructure for modern, high-growth businesses. We serve transformative, category-defining customers such as Lambda, Conde Nast, Gorgias, HeyGen, Zapier, and CodeRabbit. **About the team** Chargebee AI Labs is a Chennai-based team of AI engineers, model behaviour researchers, and deployed-AI architects. This is a research-centered team. We never assume that a foundation model, a fine-tune, or a simpler task-specific model is automatically the right solution. The team does the work of understanding the data, defining precise problems, and building strong baselines. Chargebee processes billions of payment events a year across the revenue lifecycle: credits, subscriptions, invoices, payment attempts, retries, recoveries, refunds, credit notes, disputes, and collections activity. We’re researching how this data can be used to build models that learn reusable representations of consumption, subscription and payment behaviour. **About the Role** We are looking for a Machine Learning Research Engineer who can explore new modelling approaches using subscription, usage, and payment event data. You will begin with practical prediction problems and investigate whether techniques such as self-supervised learning, sequence modelling, and multi-task learning can improve performance across multiple use cases. You don't need prior experience building a foundation model from scratch. You should, however, have experience developing and evaluating machine-learning or deep-learning models, and a strong interest in learning how reusable representations can be trained from large-scale behavioural data. This role requires both research curiosity and deep engineering rigour. You should be able to establish simple baselines, identify weaknesses in an experimental setup, implement more advanced approaches, and determine whether increased model complexity is justified. **What You Will Work On** - Data and Problem Formulation - A major part of the work will be defining the right machine-learning problem before choosing a model. - You will investigate: - How subscription, usage, and payment events should be represented - How sequences should be constructed - Which entities should define a sequence - How labels and outcomes should be created - How to handle delayed and missing outcomes - How to avoid temporal and target leakage - How behaviour differs across merchants, industries, regions, gateways, and payment methods - How usage-based and consumption-driven pricing models behave differently from flat-fee subscriptions - How models can generalise across different types of Chargebee customers - How privacy, data access, and tenant separation affect model design *Prediction and Optimisation Models* You will initially establish strong rules-based and task-specific ML baselines for problems such as: - Predicting whether a payment attempt will succeed - Selecting an appropriate retry time - Predicting whether a failed invoice is recoverable - Prioritizing collection activity - Identifying dispute risk - Predicting involuntary churn - Detecting unusual payment or subscription behaviour You will compare advanced approaches against these baselines rather than assuming that a larger model will always perform better. *Representation and Sequence Learning* Where justified by the data and business value, you will explore: - Self-supervised learning - Masked-event prediction - Next-event prediction - Contrastive learning - Multi-task learning - Temporal and sequence models - Transformer-based models - State-space models - Customer, subscription, invoice, and transaction embeddings - Transfer learning across downstream tasks - Hybrid tabular and sequential models The objective is to determine whether a shared representation can improve multiple prediction and optimisation tasks. *Evaluation and Production Validation* You will help define realistic evaluation methods for financial and transactional models. This may include: - Time-based training and test splits - Cross-merchant generalisation - Calibration and uncertainty - Class imbalance and rare-event evaluation - Offline versus online model performance - Comparison against existing rules and business processes - Model drift and changing payment behaviour - Business metrics such as payment success, recovery rate, avoided churn, and operational effort **What We Are Looking For** - Approximately 3–5 years of experience in machine learning, data science, applied research, or ML engineering. - Experience training and evaluating machine-learning or deep-learning models. - Strong understanding of statistics, probability, model evaluation, and experimental design. - Strong Python programming skills. - Practical experience with PyTorch, TensorFlow, JAX, scikit-learn, or equivalent frameworks. - Experience working with structured, temporal, behavioural, transactional, or event-stream data. - Ability to build data-processing and model-training pipelines. - Ability to establish simple baselines before introducing complex approaches. - Ability to read research papers and implement relevant ideas. - Good software-engineering and debugging skills. - Comfort working in an exploratory environment where the feasibility of an approach is not yet known. **Helpful Experience** - Sequence modelling, time-series modelling, or recommender systems - Self-supervised learning, transfer learning, or representation learning - Payments, fraud, risk, credit, advertising, marketplaces, or customer-lifecycle modelling - Survival analysis, uplift modelling, causal inference, bandits, or decision optimisation - Model calibration and uncertainty estimation - Large-scale data platforms and distributed training - Model deployment, monitoring, and experimentation - Meaningful open-source projects, publications, technical articles, or competition results **Education** A bachelor's or master's degree in comp