Harjas Romana

• Architected an high-concurrency bidding engine using React and Firebase, engineering a Redis-backed publish-subscribe pipeline to process concurrent streams and eliminate data-layer race conditions.
• Optimized a Retrieval-Augmented Generation (RAG) pipeline for an AI Voice Agent by refactoring vector query mechanisms, decreasing inference latency by 5% and improving semantic retrieval accuracy by 2%.
• Engineered a scalable, modular AI Skills Accessor utilizing Flask REST APIs and React, automating assessment generation and reducing manual evaluation workflows by 30%.