Naga Harshita Marupaka

Spearheaded the implementation of a post-inference mapping layer for apparel attribute prediction, achieving over 95% label match accuracy across 3,000+ SKUs by accurately translating model predictions into client-specific taxonomies.

Engineered and scaled multimodal Large Language Model (LLM) inference services on Kubernetes, significantly improving concurrent request handling efficiency by 20%.

Integrated advanced player telemetry and critical gameplay features for the Spurpunk game (Unity/C#), which reduced data tracking delays by 25% and accelerated performance tuning across production and test environments.

Conducted in-depth analysis of player behavior to pinpoint resource imbalances, then designed and executed targeted A/B tests that directly informed level design enhancements, resulting in a 15% improvement in player engagement.

Authored a comprehensive comparison report evaluating over 15 open-source speech recognition libraries, focusing on performance metrics, resource allocation, and conducting benchmark analysis across multiple languages to inform strategic technology decisions.

Developed and optimized a robust video transcription pipeline incorporating speaker distinction, which successfully reduced timestamp discrepancies by 20% and improved transcription accuracy.

Designed and implemented scalable REST APIs using Django Rest Framework and an interactive LIMS Dashboard with D3.js, adhering to stringent software design patterns and test-driven development principles.

Led the integration of speaker settings for a real-time group voice call feature utilizing Pub/Sub architecture, successfully piloting the solution with over 100 users.

Optimized API performance by integrating Redis caching, resulting in faster data access and a significant 60% reduction in latency from 2 seconds to 0.8 seconds.

Automated and streamlined CI/CD deployment pipelines using Cloud Build triggers, which reduced release time by 20% and enhanced development efficiency.