Research & Projects

Predict Election Outcomes via Social Media Analysis

End-to-end AI system • NLP • Data Scraping • Visualization

• Built an automated scraping pipeline using Puppeteer + TypeScript to collect large-scale Facebook post data.
• Developed NLP models in Python to classify political vs non-political content.
• Triggered secondary scraping for political posts to extract reactions, shares, comments, and engagement metrics.
• Designed a weight scoring algorithm to rank political influence using engagement and sentiment.
• Engineered a MongoDB schema (Post, FullPost, PoliticalPost) to manage processing stages.
• Created an interactive dashboard using Nuxt.js + Tailwind for trend visualization.
• Integrated everything through a Flask REST API for automation.

Tools & Techniques to Combat Cyber Radicalization (Sri Lanka)

NLP • Deep Learning • Image Analysis • Social Platforms

• Developed a radicalization detection system using NLP, deep learning, and ML.
• Collected real-world datasets from blogs, deep web sources, Twitter, Reddit, and YouTube.
• Implemented sentiment and behavior analysis for early radicalization signals.
• Built an image-based radical content detector using deep learning.
• Visualized insights using an interactive Streamlit dashboard.

Sinhala & Tamil RoBERTa Language Models

Language Models • Low-resource NLP

• Pre-trained RoBERTa models for Sinhala and Tamil using Masked Language Modeling (MLM).
• Trained on the custom-made dataset to support downstream NLP tasks.
• Focused on enabling NLP research for low-resource languages.

Asian Faces Classification Model

Computer Vision • Deep Learning • Image Classification

A deep learning–based image classification model focused on facial image analysis within Asian demographics. The project explores feature learning, dataset bias considerations, and ethical aspects of facial classification systems.

• Built and trained a convolutional neural network (CNN) for facial image classification.
• Applied preprocessing, normalization, and augmentation techniques to improve robustness.
• Evaluated model performance using accuracy and validation metrics.
• Explored ethical considerations and bias awareness in facial recognition systems.