Hursh Soni

Savitr

• Re-architected backend infrastructure for an inventory management and point-of-sale system using Terraform, reducing manual infrastructure setup time by 42%.
• Migrated backend APIs from Azure Functions to Dockerized services, and deployed them with Azure Kubernetes Service for orchestration, lowering cloud costs by 15%.
• Optimized MySQL schemas and indexes for common queries, improving read/query performance by an average of 31% in reporting and checkout workflows.
• Redesigned frontend with Next.js and shadcn/ui and used React Server Components for faster rendering.

New York University

• Developed ML models using TensorFlow and Keras to identify Big Pharma corruption (77% accuracy).
• Applied NLP techniques to analyze 1 million+ documents, uncovering multiple cases of misconduct.
• Implemented unsupervised learning algorithms, reducing false positive rates by 25%.
• Collaborated with 3 domain experts to refine analytical methods, improving overall model precision.

Z-MART

• Developed a scalable warehouse system using React, TypeScript, and Node.js, managing 10,000+ SKUs.
• Built an eCommerce checkout system using PHP, MySQL, and JS, reducing transaction fees by 63% while handling 9000+ customer transactions within the first year.
• Integrated real-time processing with socket.io, enabling instant updates for 500+ concurrent users.
• Achieved 95% code coverage through comprehensive testing with Jest and Cypress.

HealthLedger

Led the creation of an innovative blockchain-based end-to-end ecosystem revolutionizing healthcare data management by enabling secure, decentralized medical transactions and creating a transparent marketplace for healthcare services, enhancing patient privacy and cost-effectiveness. This project won first place in the healthcare track ("Best Healthcare Hack") at Georgia Tech's Hacklytics, competing against over 1,100 participants and 200 teams.

Economic Forecasting

Conducted independent research developing ARIMA and LSTM models for economic forecasting, making predictions more accurate by 30 percent. Implemented cross-validation and hyperparameter tuning, evaluating over 1,000 model configurations for optimal results. Generated visualizations of trends across 20 global economies using Matplotlib and Seaborn, with the work cited in three academic papers. Collaborated with graduate students from UCLA and UPenn to incorporate varied perspectives and advance the project.

BingAI Extension

Developed the first cross-browser extension enabling Bing's GPT-4 Chat on all Chromium browsers. Built automated Google-to-BingAI query forwarding with inline AI responses directly in search results, plus a universal popup chat. Implemented adaptive dark-mode detection with seamless theme synchronization. Scaled organically to 16,200+ active users, becoming one of the most popular early AI wrappers.