Back to Students
Luxshan Thavarasa

Luxshan Thavarasa

Software Engineer @ H2O.ai

BSc Eng Hons, Computer Science & Engineering
"Advancing AI for underrepresented languages through innovative research."

About

Luxshan Thavarasa is a Computer Science and Engineering graduate from the University of Moratuwa (BSc Eng Hons, 2025). He is currently a Software Engineer at H2O.ai, developing frontend and backend solutions for H2OGPTe and Agentic AI Applications. His research focuses on Speech Emotion Recognition (SER) for low-resource languages like Tamil, aiming to bring AI advancements to diverse linguistic communities. He is passionate about creating innovative solutions that make complex data accessible and meaningful to users.

Projects

Under the supervision of Dr. Uthayasanker Thayasivam, Luxshan made significant contributions to speech and language processing research for low-resource languages.

EmoTa: Tamil Emotional Speech Dataset

His Final Year Project focused on the Multilingual Universal Speech Emotion Recognition Model, leading to the creation of EmoTa - the first emotional speech dataset for Tamil. The dataset contains 936 utterances from 22 native Tamil speakers across 5 emotions (anger, happiness, sadness, fear, neutral), achieving F1-scores of 0.91 with XGBoost.

📄 Paper: CHiPSAL @ COLING 2025  |  💻 GitHub Repository

DravidaKavacham: Abusive Content Detection

Developed an open-source tool for detecting abusive content in Tamil and Malayalam code-mixed text, achieving 0.78 F1 on Tamil and 0.70 F1 on Malayalam using transfer learning and multi-head attention mechanisms.

💻 DravidaKavacham GitHub  |  Accepted at DravidianLangTech @ NAACL 2025

Multilingual Universal SER Model

Trained a universal speech emotion recognition model supporting 15+ languages using Wav2Vec2 and XLS-R architectures.

Research Areas

Speech Emotion Recognition • Natural Language Processing • Low-Resource Language Processing • Multilingual AI • Agentic AI Systems

Technical Skills

Python, PyTorch, TensorFlow, TypeScript, React, FastAPI, AWS, Docker, Speech Processing, NLP

Key Notes

  • Software Engineer at H2O.ai (2024-Present)
  • Publication: EmoTa - Tamil Emotional Speech Dataset (COLING 2025)
  • Publication: DravidaKavacham - Abusive Text Detection (NAACL 2025)
  • Publication: Global PIQA - Commonsense Reasoning Benchmark (arXiv 2024)
  • All Island Mathematics Competition 2018: 2nd Runner-Up (Northern Province)