Rithvik Thota

Architected and deployed a production-grade, kiosk-based cleanroom access control system using Microsoft Access and VBA, integrating hardware card swipe input for real-time user authentication. Designed a stateful session management system handling check-in, check-out, forced closure, and automated end-of-day reconciliation, ensuring data consistency and eliminating stale sessions Built a low-latency input processing pipeline for magnetic stripe readers with debounce protection and buffered parsing for reliable swipe handling.Developed an admin control dashboard with real-time occupancy tracking and manual override capabilities, improving operational visibility and control Implemented a comprehensive audit logging and reporting system, including dynamic filtering and Excel export, enabling traceability and supporting billing workflows. Refactored and merged independent system modules into a unified architecture, resolving dependencies and improving system stability and portability

Designed and implemented scalable machine learning pipelines for predictive analytics tasks, leveraging Python, TensorFlow, and data processing frameworks. Optimized data ingestion, preprocessing, and model deployment workflows, resulting in a 40% reduction in overall data processing time and improved system efficiency for large-scale datasets. Fine-tuned and optimized state-of-the-art Generative AI models, including transformer-based architectures, for enhanced performance on domain-specific tasks. Applied techniques such as transfer learning, hyperparameter tuning, and mixed-precision training, leading to a 25% boost in model accuracy and a 20% reduction in training time, significantly improving inference quality and resource utilization.

Architected scalable and high-throughput data pipelines to handle real-time and batch atmospheric data from satellites and IoT-based weather sensors. Utilized distributed computing frameworks to ensure improved system reliability, scalability, and data throughput under varying data volumes. Built, validated, and optimized machine learning models for atmospheric event prediction using scikit-learn, XGBoost, and GridSearchCV, improving alert accuracy and reducing false positives. Collaborated cross-functionally with data scientists, engineers, and climate researchers to deploy predictive models into production environments, ensuring alignment with scientific standards and operational efficiency. Proactively identified bottlenecks and proposed long-term solutions for sustainable climate data infrastructure. Utilized NARL HPC Cluster an indigenous high-performance computing system for scalable model experimentation and secure, policy-compliant climate data processing. Co-authored technical documentation and contributed to published research in IEEE journals, focusing on storm nowcasting models powered by GNSS-based machine learning.

Generated actionable statistical insights by analyzing large-scale environmental and climate datasets using Python, pandas, and statistical modeling techniques. Partnered with environmental scientists and domain experts to refine prediction models, leading to a 20% improvement in forecast accuracy for temperature and precipitation trends which are impacting climate policy and strategic planning efforts. Redesigned and automated data analysis workflows, transitioning from time-intensive manual processes to efficient script-based pipelines. Leveraged Jupyter Notebooks, NumPy, and Bash scripting to automate data cleaning, transformation, and reporting tasks, cutting processing time from over 2 hours to just 30 minutes and enabling faster data-driven decisions.