Built Orbis - a multilingual classification system for 1.4M+ global corporate entities across 120+ countries, combining fine-tuned XLM-RoBERTa, deterministic rules, and async batch inference at 98.75% precision

Improved pipeline throughput 80% via parallel execution, memory-efficient batching, and profiling-driven SQL index optimization on large structured datasets

Replaced months of manual labeling by building an async inference pipeline processing 500K+ multilingual entities across 50+ languages

Owned RESTful backend services in Node.js on AWS handling real-time user traffic - cut synchronization latency 60% through API refactoring and indexed query optimization

Shipped OAuth2 authentication and CI/CD pipelines enabling zero-downtime weekly deployments to production

Developed and deployed NLP inference backend integrating BERT-based models, improving intent recognition accuracy by ~25%. Optimized request handling and backend performance to reduce API latency under concurrent usage.

Built Python ML scripts for student-facing data science curriculum at an AWS-backed EdTech platform serving 800+ colleges.