Machine Learning Engineer

About the Role

You will build ML systems that improve payment success rates, detect fraud in real-time, and automate merchant risk assessment. Every 1% improvement in success rate translates to crores in additional GMV for our merchants. 

You'll own the full ML lifecycle: feature engineering, model development, deployment, and monitoring. This is a greenfield role—you'll shape the ML architecture from scratch. 

Roles & Responsibilities 

1. Build smart payment routing models to optimize transaction success rates across UPI, cards, and net banking rails 
2. Design and deploy real-time fraud detection systems with sub-100ms latency—device fingerprinting, behavioral analysis, velocity checks 
3. Develop merchant risk scoring models for automated underwriting using documents, transaction patterns, and external signals 
4. Build document intelligence pipelines—OCR, classification, and data extraction for KYC documents (PAN, Aadhaar, GST, bank statements) 
5. Set up ML infrastructure—feature stores, model serving, A/B testing frameworks, monitoring and alerting 
6. Collaborate with product and engineering teams to integrate ML models into production systems 
7. Monitor model performance, detect drift, and implement retraining pipelines 
8. Document model architecture, training procedures, and performance metrics  

Requirements 

1. 3-5 years of applied ML with models deployed to production (not just notebooks/Kaggle) 
2. Strong Python with PyTorch or TensorFlow, plus pandas/numpy for data manipulation 
3. End-to-end ML skills: feature engineering, training, evaluation, deployment, monitoring 
4. Experience with tabular/transactional data and classification/ranking problems
5. Understanding of real-time inference—latency budgets, feature stores, model serving 

Good to Have 

1. Fraud detection or risk modeling experience in fintech/payments 
2. Multi-armed bandits or reinforcement learning for optimization 
3. Graph neural networks for network-based detection 
4. LLM experience: RAG pipelines, fine-tuning, prompt engineering