ROLE OVERVIEW

Role includes the development and application of transformer-based foundation models that learn from large-scale customer-event sequences. The role combines deep technical ownership with project leadership: shaping modelling strategy, guiding junior team members, managing stakeholder expectations and ensuring that solutions are validated, explainable and ready for governed implementation. The successful candidate will remain hands-on while setting standards for model quality and delivery.

KEY RESPONSIBILITIES

- Define the modelling roadmap for self-supervised pretraining, representation learning and downstream adaptation across fraud, delinquency, acquisition and behavioural-risk use cases.
- Lead architecture choices, training strategy, evaluation design and benchmarking against strong classical ML baselines.
- Guide approaches for multi-label learning, task-specific heads, attention pooling, embeddings and fine-tuning while controlling leakage and overfitting.
- Own end-to-end model validation, including out-of-time performance, stability, drift, explainability and limitations.
- Translate business objectives into analytical workplans, milestones, resource requirements, risks and measurable success criteria.
- Partner with ML engineering, data, cloud, model validation and business teams to move models from research into controlled implementation.
- Communicate technical trade-offs and results to senior stakeholders; prepare governance-ready methodology, validation responses and decision materials.
- Monitor delivery against plan, resolve dependencies and escalate material risks with clear mitigation options.

REQUIRED SKILLS AND EXPERIENCE Programming: Strong proficiency in Python and SQL; practical experience with NumPy,



Pandas and version-controlled development using Git

Deep learning: Hands-on experience with PyTorch and transformer architectures; understanding of attention, embeddings, self-supervised learning, transfer learning and fine-tuning

Modelling: Strong foundations in supervised learning and benchmarking using XGBoost, LightGBM, scikit-learn or logistic regression

Validation: Experience with class imbalance, out-of-time validation, stability and drift; familiarity with metrics such as Gini, KS, PR-AUC and capture rate

Data: Experience converting categorical, numeric and temporal event data into leakage-controlled, model-ready datasets; PySpark or distributed processing exposure

Ways of working: Transparent communication, structured problem solving, documentation discipline and ability to collaborate with engineering, validation and business stakeholders

GOOD TO HAVE

- Credit risk, fraud, collections, underwriting or financial-services analytics experience.
- Experience building foundation models for tabular, behavioural, recommendation or event data rather than only text.
- Large-scale distributed training, model compression, serving optimisation or real-time inference.
- Model risk governance, independent validation interaction or deployment in a regulated environment.

QUALIFICATIONS

- Bachelor’s or Master’s degree in Data Science, Statistics, Mathematics, Physics, Computer Science, Engineering or another quantitative field; advanced research qualifications are welcome.
- 4–8 years of applied machine-learning experience with strong hands-on depth in deep learning and transformers.

Demonstrated ownership of complex analytical deliveries and leadership of technical teams or workstreams.

📌 Senior Data Scientist (Gurugram)
🏢 EXL
📍 Gurugram