Praveen Kumar
I am a Machine Learning Engineer with over 6 years of experience creating robust and scalable machine learning pipelines and deep learning solutions across diverse industries.
About
I am a Machine Learning Engineer with over 6 years of experience creating robust and scalable machine learning pipelines and deep learning solutions across diverse industries. I specialize in designing end-to-end data workflows, large language model training and fine-tuning using TensorFlow, PyTorch, and Hugging Face, and optimizing pipelines on cloud platforms such as AWS and Snowflake. I excel at translating complex business challenges into impactful, production-ready AI systems.
Throughout my career, I have developed expertise in MLOps, natural language processing, big data tools, and cloud-native infrastructure, consistently improving model accuracy, reducing latency, and automating deployment workflows. I am passionate about leveraging cutting-edge AI technologies and data engineering best practices to build scalable, efficient, and secure AI-driven applications that deliver measurable business value.
- 6 years' experience
- Joined August 2025
Industry experience
- Financial Services
- Healthcare
- Software & Internet
- Retail
- Professional Services
Work experience· 6 roles
At eBay, I engineered scalable ML pipelines using PySpark and Airflow for real-time clickstream data processing, improving product recommendation accuracy by 22%. I optimized PyTorch model serving with TorchServe, doubling API response speed and reducing cloud resource costs by 18%. I enhanced ETL processes and accelerated training cycles using Pandas and Spark. Integrated NLP capabilities leveraging spaCy for named entity recognition and topic modeling, boosting user insight extraction by 60%. Containerized models via Docker and Kubernetes to reduce inference latency by 35%. Implemented mixed precision training with PyTorch Lightning, speeding up training by 40% and improving accuracy. Built real-time monitoring and anomaly detection systems using Kafka and Spark Streaming, increasing detection coverage by 60%. Integrated FAISS vector search for improved product metadata lookup, raising search relevancy by 18%. Automated model retraining workflows with MLflow and SageMaker, cutting ma
Engineered scalable ML pipelines using PySpark and Airflow, improving product recommendation accuracy by 22%. Deployed PyTorch models with TorchServe, doubling API response speed and reducing cloud costs by 18%. Enhanced ETL and training data readiness by 3x using Pandas and Spark. Integrated spaCy NLP components to improve user insights by 60%. Reduced inference latency by 35% through containerized deployments on Kubernetes-managed GPU clusters. Accelerated training speed by 40% using PyTorch Lightning with mixed precision. Built real-time monitoring systems with Kafka and Spark Streaming, increasing anomaly detection coverage by 60%. Improved search relevancy by integrating FAISS vector search. Automated retraining workflows with MLflow and SageMaker, cutting downtime by 70%. Developed a RAG-based low-latency recommendation system, increasing upsell conversions by 15%. Modernized deployment pipelines, reducing infrastructure costs by 25%. Migrated analytics to AWS EKS and Terraform,
At Evernorth, I built an XGBoost fraud detection model identifying $3M in suspicious claims, enhancing financial risk management. Developed an LLM-powered Q&A system boosting query resolution by 50% using LangChain and Hugging Face. Created AI fraud detection pipelines using Isolation Forest and Autoencoder reducing annual financial loss by $1M. Improved customer churn prediction accuracy by 22% with SHAP features and ensemble models. Automated training workflows with Airflow and MLflow, reducing manual work by 40%. Used BERT for NLP pipelines that enhanced ICD-10 coding accuracy by 27%. Decreased HEDIS penalties by 19% employing LightGBM on longitudinal patient data. Processed 10TB patient data weekly with Apache Spark and Synapse. Automated HIPAA-compliant deployments with AWS SageMaker, reducing overhead by 60%. Improved early chronic disease detection by 21% via LSTM models. Implemented observability with Prometheus and Grafana to guarantee 100% SLA compliance. Designed synthetic d
Education
Certifications and awards· 6
Hire Praveen Kumar today
Post a job and invite Praveen, or contact them directly. No hidden commission.







