Shrihari Athreyas

I’m a Computer Science (Data Science) undergraduate with a systems-oriented, data-driven, research-focused approach to problem solving. I focus on end-to-end development, especially data curation with messy, heterogeneous datasets, where I analyze distributions, class imbalance, and feature relevance to guide model design.

In my projects, I’ve built high-performing and testable ML systems. AksharaVision used a Swin Transformer for Kannada handwritten character recognition, curating a 113-class imbalanced dataset and expanding 5.6k to 40k+ samples—achieving 99.89% accuracy and 0.9989 macro-F1 with robustness testing and error analysis. AttriShield improved employee attrition prediction from 86% to 92.5% accuracy using SMOTE and threshold tuning, and CarnaticPrep delivered a RAG pipeline over 209 PDFs + 58K CSV rows with structure-aware chunking and CrossEncoder reranking for grounded, context-aware answers.

I also bring practical experience from an internship in Android app development using Gen AI, where I worked on Firestore schema planning, backend workflow understanding, recommendation system concepts, and AI-assisted product engineering using Google AI Studio and generative AI tools. I value iterative, experiment-driven development—diagnosing failures, improving reliability, and designing for real-world generalization.