Senior Data Scientist / Machine Learning Engineer

● Implemented LLM-driven features for a personal CRM, including smart notifications, RAG-based action recommendations and predictive models for user retention and personalized communications, increasing subscriptions by 32%.

● Owned a production XGBoost-based ML system to predict real-time availability across millions of products, replacing unreliable store-reported stock status and reducing cancelled orders by 12%. ● Upgraded the delivery time prediction system with a multi-stage LightGBM architecture and automated CI/CD pipelines for training and real-time inference, improving ETA error by 25% over the previous version. ● Built an internal LLM-powered Text-to-SQL chatbot using RAG and schema-aware prompting, cutting data-related Jira requests by 80% and enabling self-service reporting for business teams.

● Owned a real-time wildfire detection system using YOLO, CNNs, and OpenCV , including model training on field data, video processing, evaluation, and automated alerts within 30 seconds, achieving 96% accuracy. ● Developed time-series forecasting and anomaly detection solutions for air-quality and energy platforms, using LSTM, ARIMA, Prophet and other ML methods to predict PM levels, electricity consumption, and detect abnormal patterns. ● Designed NLP pipelines for text classification and knowledge graph construction from unstructured nutrition data, using document embeddings and entity recognition to reduce processing time by 10x. ● Led a 4-person team, guiding task planning and technical needs to deliver ML solutions meeting business requirements.

● Developed Python ML libraries with FPGA/GPU acceleration, achieving up to 20x speedups. ● Built and evaluated end-to-end ML solutions, from statistical models for tabular data to neural networks for CV/NLP, with Dockerized cloud deployments for scalable inference. ● Led our system's successful submission to MLPerf benchmark, collaborating with companies including Google and NVIDIA. ● Consulted with global fintech clients to deliver data-driven insights for credit scoring, fraud detection and stock prediction.

● Automated database entry and enabled large-scale geospatial data analysis through visualization and statistical analysis ● Improved event detection efficiency by 42% compared to traditional methods, by building machine learning techniques

● Developed a custom data analysis tool for medical information, enabling efficient trend detection and data visualization