Hi, I'm Dasith 
AI & Machine Learning Enthusiast. Software Engineer at IFS & Master's Student at Curtin University. I document life outside the terminal on Instagram & TikTok.
→ Tap for my study materials
About
I'm currently pursuing a Master's in Artificial Intelligence at Curtin University. My focus is on machine learning, data-driven systems, and real-world AI applications. Beyond my studies, I work as a Software Engineer at IFS and spend time building AI-powered projects. These projects allow me to explore advanced AI techniques while solving practical problems.
Recent Posts
View all →
Work Experience
Education
Dean's List, Semester 1: 90 average across all units
Skills
Machine Learning / AI
Languages
English
SinhalaI like building models
During my time in university, I completed multiple assignments focused on ML and AI. Each project was an opportunity to train models, experiment with algorithms, and apply AI techniques to solve real problems.
October 2026
PAM-E: Escalation-Aware Context Transfer Between AI Agents
Curtin University
Designed and built PAM-E, a protocol that decides what an AI support agent must hand over when it escalates a case, carried inside a signed Portable Agent Memory (PAM) artefact. Tested against generic PAM with matched runs, a live L1 to L2 to L3 help desk and all 10,042 chats of the public ABCD dataset, where PAM-E kept 83.6% of the labelled facts against 37.7% for PAM at the same size. Explore the workspace, the research dashboard and every recorded test.
October 2026
SustainaBuild: Reinforcement Learning for Building Energy Control
Curtin University
Compared DQN, REINFORCE and A2C for joint cooling and battery decisions in a simulated Perth office. Historical weather, held-out evaluation and conventional benchmarks reveal the trade-offs between cost and comfort. Explore the interactive energy lab and recorded research evidence.
June 2026
Explainable Multi-label ECG Diagnosis for Cardiovascular Conditions
Curtin University
Built an explainable classifier for 12-lead ECGs (NORM, AFIB, AFLT, 1dAVb, RBBB and LBBB) using a 1D ResNet on raw PTB-XL waveforms, with condition-specific clinical evidence checks for rhythm, PR interval and bundle-branch morphology, and a review pathway for uncertain cases. Reached a macro AUROC of 0.986 on the held-out test fold and 0.973 on external Chapman-Shaoxing data.
May 2026
Suicide Risk Detection from Clinical Conversations Using Hybrid LLMs
Curtin University
Benchmarked MentalBERT, a prompted GPT-4o-mini, a Chain-of-Thought LLM and a hybrid encoder and LLM design for five-level risk classification aligned with the Columbia Suicide Severity Rating Scale. The hybrid reached 0.703 accuracy on an 808-case held-out set, with auditable reasoning, a clinician-facing dashboard and Privacy Act 1988 compliance built in.
March 2026
Domain Adaptation for ADAS Perception in Australian Freight Environments
Curtin University
Benchmarked state-of-the-art segmentation models (SegFormer, Mask2Former, OneFormer, YOLOv8) on the SydneyScapes dataset. Quantified domain shift from Cityscapes, applied fine-tuning achieving +15.81pp mIoU improvement. Focused on harsh Australian conditions: night, dust, rural roads, and wildlife.
2025
Enhancing Electric Vehicle Battery Health Forecasting Using Advanced ML
University of Westminster
Built a multi-model AI pipeline to predict battery State of Health (SoH) and cycle life. Combined lab and real-world usage data with transfer learning. Applied KNN imputation, Isolation Forest, SMOTE, and SHAP for explainability using Random Forest and XGBoost.
2025
AI-Based Fruit Detection and Assistance System for Visually Impaired
University of Westminster
Designed a real-time fruit detection system to assist visually impaired individuals during everyday tasks like shopping. Used YOLO-based object detection with OpenCV, converting detection results into audio feedback for accessibility.
Study Materials
Comprehensive notes from my AI & ML journey. Battle-tested materials I created and refined through my studies.
Search & Logic Approaches in ML
11 topics • Updated Oct 2025
Interactive lecture companions covering search, logic and reasoning. Open the module to browse each topic.
Planning & Handling Uncertainty in Machine Learning
11 topics • Updated Oct 2026
Interactive lecture companions on probabilistic reasoning, Bayesian networks, temporal models, generative models and planning under uncertainty. Open the module to browse each topic.
Or explore other notes here
Get in Touch
Reach out on TikTok or Instagram, I'd love to connect!
For work inquiries, reach me directly at dasithcg@gmail.com