Developed a scalable Amazon product scraper in Python by containerizing Playwright in Docker with parallel execution, reducing validation workflows from days to under 30 minutes and cutting data mismatches by 50%

Designed a Purchase Order matching module combining internal MSSQL and customer Postgres data, enabling accurate audits, inventory control, and sales order management across company and factory POs

Created an interactive forecasting dashboard in Next.js and React consolidating 10+ API outputs on purchasing trends, inventory projections, and purchase orders, enabling CFO and e-commerce team to optimize planning and ordering

Implemented a CLIP-based AI image comparison pipeline that detects mismatches across 1800+ Amazon and internal product images, replacing broad manual reviews with targeted analysis and improving product page quality

Architected a Docker Compose and NGINX deployment framework isolating dev and production services in a shared network, streamlining deployments and enhancing environment security

Automated ticketing workflows in legacy ERP system, significantly reducing manual entry by 90%, improving communication with production team, and boosting packaging throughput

Achieved 100% overhead cost reduction and significantly boosted performance by reengineering NEO as a VSCode extension, eliminating AWS server infrastructure and Docker

Upgraded developer experience with interactive React Vite Webview panel, VSCode API, and custom React chart components for a clean and intuitive interface

Optimized SEO analysis by using Puppeteer to access local instances and navigate pages, leveraging Performance API for extracting 6 core performance metrics

Enhanced data accuracy and usability by designing an algorithm that transforms raw performance data into scores

Enforced static type error handling with TypeScript ensuring scalability, code readability, and minimized runtime errors

Improved performance and render speed by ~64% using Recharts and a custom useMetrics state hook, eliminating previously needed useEffect and Canvas API

Engineered intuitive spatial audio feedback protocol with 90.4% user accuracy using MATLAB, which converts single-point LiDAR camera distance data transmitted via Arduino into comprehensible frequencies relating to distance

Designed testing environment to assess feedback intuition and collect user data while navigating obstacle course blindfolded

Analyzed system integration data using Python, Pandas, and Seaborn to show ~2.5x improvement in user’s obstacle avoidance

Conducted in-depth literature review regarding existing visual mobility aides, visually impaired navigation, electronic white canes, and object detection algorithms