Rajesh Kumar Sharma

Professional Experience

Personal Sabbatical

Aug 2025 - Present

Focused on GenAI certifications and hands-on systems across RAG, MCP, LLMOps, database agents, GenAI applications, reinforcement learning from human feedback, fine-tuning, and AI engineering.

Flipkart Internet Pvt Ltd - Senior Data Scientist

Oct 2020 - Jul 2025 | Bangalore, India

  • Improved payment gateway success rate by 84 bps absolute, driving $1.1M annual savings and $30M incremental revenue.
  • Built batch and realtime credit risk models for 1P EMI with 11% relative credit-loss reduction and 41% whitelist increase using graph embeddings, sequential features, domain adaptation, user activity signals, and Experian profiles.
  • Built realtime card scanning ML system outperforming OCR baselines, with synthetic image generation, EfficientNet, YOLO, FastAPI deployment on Azure, and model-size compression from 245 MB to 1.2 MB.
  • Improved settlement models by 31% (relative), driving around $75K/month incremental recoveries.
  • Estimated credit hungriness, credit bureau score, income, and employment status for a 300Mn+ active-customer base using multitask learning.
  • Reduced 3000 features to 128-dimensional embeddings using a transformer-based autoencoder, cutting training and inference time while maintaining model performance.
  • Built SMS-based EMI risk model using CBOW embeddings of SMS sender entities, reducing inference latency from 500 ms to 10 ms and improving ROC-AUC by 80 bps.
  • Mentored senior and junior data scientists on modeling, documentation, deployment, non-functional requirements, terabyte-scale data handling, and GCP workflows.

Lemnisk Pvt Ltd, formerly Vizury Interactive Solutions Pvt Ltd - Senior Data Scientist

Jul 2018 - Sep 2020 | Bangalore, India

  • Built deep reinforcement learning model with 30% lift in clicks for flagship product Ramanujan.
  • Delivered recommendation MVP end-to-end and deployed with Flask + Docker on AWS.
  • Built anomaly detection pipelines to detect bot traffic and improve data reliability.
  • Explored AWS SageMaker, word embeddings as features, and transformer architectures for production use cases.
  • Led hiring and mentoring, and served as acting manager for two direct reports.

Earlier Roles

  • Lemnisk Pvt Ltd - Data Scientist (Jul 2016 - Jun 2018): built deep neural networks for click-through-rate and conversion-rate prediction with automated preprocessing, training, hyper-parameter tuning, deployment, and monitoring; increased re-targeting profit by 35%.
  • Neo Metrics, acquired by Accenture Spain - Fellow / Junior Consultant (Dec 2010 - May 2012): estimated social-media ad-spend efficiency and simulated energy setups to reduce cost and environmental impact while meeting demand.
  • Indian Institute of Technology Kanpur - Research Associate (May 2010 - Nov 2010): modeled inter-speaker variability in automatic speaker recognition.