Projects
Sysco Pulse: Graph RAG-Based Analytical Chatbot
Graph RAG-based analytical chatbot for customer, vendor, and stock analytics with table generation, visualization (Plotly), and reasoning capabilities using GPT-4.1 and Neo4j. Features include dynamic table generation from queries, interactive Plotly visualizations, and multi-hop reasoning over knowledge graphs. Achieved Second Runner-Up at Codefest 2024 (SLIIT, Sri Lanka).
Technologies: Graph RAG, Neo4j, GPT-4.1, Plotly, Python, Knowledge Graphs, NLP
Monocular Depth Estimation Case Study
Case study on monocular depth estimation using KITTI and Virtual KITTI datasets. Replicated Eigen network and Depth Prediction Transformer models for benchmarking under standard and adverse conditions (rain, fog, low-light scenarios). Includes comprehensive evaluation on KITTI Eigen split with standard metrics.
Technologies: PyTorch, KITTI, Transformers, Computer Vision, Deep Learning
SafeGuard: Edge AI-Based Construction Safety Monitoring System
Edge AI system for construction site entrance featuring facial recognition, personal protective equipment (PPE) detection, and health monitoring on Raspberry Pi. Enables automated worker check-in and real-time safety compliance monitoring. Designed for on-device inference to minimize latency and privacy concerns.
Technologies: Raspberry Pi, Computer Vision, Edge AI, TensorFlow, Facial Recognition
FrauxAI: Machine Learning-Based Fraud Detection for Mobile Banking
Machine learning models for payment fraud detection with facial recognition-based login. Integrated as a mobile banking SDK for deployment in Sri Lankan financial applications. Combines transaction pattern analysis, behavioral biometrics, and facial authentication.
Technologies: Machine Learning, Facial Recognition, Mobile SDK, TensorFlow, Scikit-learn
Financial and Non-Financial Data Analytical Chatbot
Conversational and analytical chatbot for financial and non-financial data analysis. Built using LB Finance Sri Lanka annual report integration with analytical query capabilities and visualization support. Processes both structured and unstructured data from annual reports.
Technologies: NLP, LLMs, Data Analytics, Python, Financial Analysis
Cranium Cryptics: Brain Disease Detection Application
Web application for detecting tumor, Alzheimer's, and ischemic stroke using CNN models (VGG16, VGG19, ResNet50, EfficientNet) with Flask backend. Provides medical professionals with automated diagnostic support through medical image analysis.
Technologies: CNN, Flask, TensorFlow, Medical AI, Medical Image Analysis
Adaptive Genetic Algorithm for Vehicle Routing Problem with Time Windows
Adaptive genetic algorithm with machine learning-based parameter control for vehicle routing problem with time windows (VRPTW). Optimizes routes under time window and capacity constraints. Suitable for real-world logistics and delivery optimization applications.
Technologies: Genetic Algorithm, Optimization, Machine Learning, Operations Research