U.Ajay Kumar

End-to-End RAG-Based Analytics Chatbot

• Developed a production-grade RAG chatbot using OpenAI + open-source LLMs, LangChain, LangGraph, and FAISS.
• Enabled natural language querying over API docs, help center content, and analytics data.
• Integrated with Snowflake and dynamic SQL generation for real-time querying.
• Deployed using Flask backend on AWS ECS, with containerized, scalable architecture.

ETA/ETD Prediction – Logistics Optimization

• Built ETA/ETD prediction models using XGBoost and Random Forest on real-time logistics data.
• Achieved 85%+ accuracy, improving delivery forecasting and operational planning.
• Processed 20M+ transport records including route history, delays, weather, and distances.
• Deployed models on EKS/GKE, leveraging SQL pipelines for feature engineering.

Delivery Assurance Prediction Model

• Trained Deep Learning and CatBoost models to predict delivery success with 90%+ accuracy.
• Handled 20M+ records with advanced preprocessing and feature selection.
• Served via Flask APIs, deployed on AWS SageMaker and Kubernetes for real-time inference.
• Implemented model drift monitoring, regular fine-tuning, and continuous improvement cycles.

Smart Provider Recommendation Engine

• Engineered a recommendation model to suggest optimal providers based on reliability, cost, and cancellations.
• Achieved 85%+ accuracy using neural architectures and CatBoost.
• Deployed on Docker + Kubernetes, integrating LangGraph for decision logic orchestration.
• Designed real-time ML pipelines using SQL for processing 20M+ operational records.

Game Recommendation Engine Project Overview Key Achievements Tech Stack & Modeling Deployment & Integration Frontend & Visualization

Role: Data Scientist / Big Data Engineer
Organization: BizAcuity

Designed and deployed a high-impact, production-grade recommendation engine for new, popular, and continuing games.

• Achieved 80% mean click-through rate (CTR) by leveraging user behavior data and personalized recommendations.
• Boosted user engagement and retention through highly accurate suggestions.
• Utilized Bayesian Optimization and Hyperopt for hyperparameter tuning and model refinement.
• Fine-tuned models through systematic experimentation for optimal performance.
• Developed core logic in Java and exposed services via Spring Boot APIs.
• Deployed as a production-ready system, integrated seamlessly across multiple systems.
• Ensured scalability and reliability for real-world usage.
• Contributed to frontend UX improvements, ensuring a seamless user experience.
• Built interactive dashboards to monitor KPIs and visualize large-scale recommendation analytics in real-time.

Churn Prediction Model – Telecom Industry

• Built a churn prediction model with 85%+ accuracy to identify at-risk customers.
• Applied SMOTE to address class imbalance and enhance model performance.
• Used Snowflake for scalable data processing and feature engineering.
• Developed REST APIs to serve real-time churn predictions.
• Deployed model using Docker on AWS ECS for scalable integration.
• Delivered actionable insights for proactive retention strategies.