Puneet Singh

About Puneet Singh

Machine Learning Engineer with 5 years of experience building and deploying production ML systems in banking and financial services.1. My core work revolves around problems that have real financial consequences — fraud detection, credit risk modelling, and NLP on unstructured banking data. I have worked across the full model lifecycle, from data engineering and feature development to deployment and post-production monitoring. A model that drifts quietly in production without anyone noticing is the kind of failure I take personally.2. A big part of what I do is the engineering that sits between a well-performing model and a reliable production system — retraining pipelines, drift detection, model interpretability, and making sure the system holds up under real-world conditions. This is where I have spent most of my technical energy and where I think the real depth of ML engineering lives.3. My day-to-day stack: Python, SQL, PySpark, Scikit-learn, XGBoost, CatBoost, BERT, MLflow, HiveQL. I have worked with both classical ML and deep learning approaches, and I pick based on what the problem actually needs — not what is trending.4. Over the past year I have been building seriously in the GenAI space — RAG pipelines, multi-agent systems, and LLM-powered data tools using LangChain, LangGraph, Pinecone, and CrewAI. It is the fastest-moving part of the field right now and I intend to stay close to it.Always open to connecting with engineers and practitioners working on serious ML problems.

Experience

1. Consultant Ii

   EXL

   Nov 2025 — Present · US

   Used LLM to generate contextual text embeddings from business data, capturing semantic meanings and these embeddings formed the core input features for a multi-class classification model.• Built a hybrid ML pipeline combining LLM-based embeddings with Boosting technique, solving the business problem by leveraging the strengths of both deep language understanding and gradient boosting on structured features.• Enhanced the model through advanced feature engineering and interpretability techniques — identifying which features drive predictions and making the model\'s decisions more transparent for business stakeholders.