Associate Fraud and Federated AI

The opportunity You will be part of NPCI’s Market Innovation team, working at the intersection of advanced machine learning, deep learning, graph AI, and Generative AI to build next-generation intelligent systems for India’s digital payments ecosystem. This role focuses on solving India-scale problems such as fraud detection, mule/AML risk modeling, transaction intelligence, and conversational AI, using both classical ML and cutting-edge AI architectures (LLMs, GNNs, Transformers).

You will design end-to-end AI systems—from problem formulation, feature engineering, and model development to GPU-accelerated optimization and production deployment, ensuring low latency, scalability, and robustness. The role offers a unique opportunity to work on: Graph-based fraud detection systems LLM-powered platforms (RAG workflows) GPU/CUDA optimized AI pipelines Privacy-preserving and federated AI systems You will collaborate with top academic institutions (IITs/IISc) and cross-functional teams to push the boundaries of applied AI in financial systems.

Job details Job Title: Data Scientist

• Associate Fraud and Federated AI Division: NPCI Market Innovation Experience: 1 to 3 Years Education: B.Tech / M.Tech / MSc / MCA (PhD preferred) in CS, AI, DS, Mathematics or related field Employment Type: Full-time Location: Mumbai & Hyderabad Role Type: Permanent Key responsibilities Machine Learning & Advanced Modeling Develop and deploy ML/DL models (Logistic Regression, RF, XGBoost, NN, CNN, Transformers, GANs) Build models for fraud detection, AML, anomaly detection, transaction intelligence Work on imbalanced datasets using advanced sampling and cost-sensitive learning Graph AI & Advanced Systems Design Graph AI models: GNN, GCN, GAT, temporal graph networks Apply network analytics for fraud rings, mule detection, behavioral risk signals Generative AI Build LLM-powered applications (chatbots, complaint intelligence, document analysis) Implement: RAG pipelines Prompt engineering & LLM fine-tuning Feature Engineering & Data Science Perform EDA, feature engineering (temporal, behavioral, aggregated features) Work with structured, semi-structured, and unstructured data Model Optimization & GPU Acceleration Optimize models for: Latency & throughput GPU performance (CUDA-based optimization) Use libraries such as: RAPIDS, cuDF, cuML, cu. Graph, Py. Torch Geometric Evaluation & Experimentation Design custom loss functions (weighted BCE, cost-sensitive) Apply business-aligned metrics: Precision@K, Recall, ROC-AUC, PR-AUC Use robust validation techniques (cross-validation, time-based splits) Deployment & Production Systems Integrate models into batch and real-time production systems Design scalable ML pipelines & APIs Monitor: Model drift Performance stability Business impact Collaboration & Research Work with data engineers, product teams, and business stakeholders Contribute to research, innovation, and academic collaborations Stay updated on latest AI advancements (LLMs, Graph AI, Federated Learning) Requirements Required Technical Skills Core ML & Data Science Strong in: Supervised & unsupervised learning Statistical modeling (Logistic Regression, DA) Tree models (RF, XGBoost, LightGBM) Deep Learning: NN, CNN, Transformers, GANs Generative AI & LLM Stack Hands-on experience with: LLMs (OpenAI, open-source models) Prompt engineering, fine-tuning RAG pipelines & vector databases Graph AI Experience with: GNN, GCN, GAT Graph-based fraud detection Network analytics Programming & Tools Strong proficiency in: Python (Num. Py, Pandas, scikit-learn) SQL (large-scale data processing) Frameworks: Py. Torch / Tensor. Flow Py. Torch Geometric Good to have skills and experience required Experience in: Payments / fintech / banking domain Fraud detection, AML, mule detection systems Exposure to: Graph analytics on transactional data Federated learning & privacy-preserving AI Real-time streaming systems Experience with: Cloud platforms (AWS/GCP/Azure) ML pipelines & MLOps frameworks Research experience: Publications in ML/AI conferences or journals Ability to: Design AI models inspired by mathematics/physics principles