Research Interests
Areas where I want to focus my doctoral research, with the concrete concepts and tools I'd be working from.
I want my research to make a tangible difference for the people who need it most. Every project I've built so far has traced that motivation: CoinVision put a Sri Lankan coin recognition model onto affordable edge hardware so children in low-resource contexts could build early savings habits from small change; EngageTrack framed classroom engagement as a computer vision problem so remote learners wouldn't stay invisible to their teachers; and the Interactive Money Saver we presented at TIBS 2026 pulled both threads together. Computer vision, edge AI, and ML for education and HCI aren't abstract interests for me — they're the areas where a working model most directly changes what someone can do with their day. What I'd want from a PhD: to keep working on problems where a deployed model can help real people, and to build the theoretical rigour to know when the model is actually helping.
Computer Vision & Deep Learning
CNN architectures, video understanding, anomaly detection, and dataset curation for under-represented visual domains. Grounded in hands-on work with TensorFlow, OpenCV, and custom-trained models.
- CNNs
- Video Understanding
- Anomaly Detection
- Dataset Curation
ML for Education & HCI
Modelling learner engagement from multimodal signals; applying computer vision to virtual-classroom problems. Framing pedagogical questions as tractable ML problems.
- Engagement Modelling
- Multimodal ML
- Virtual Classrooms
IoT & Edge AI
Deploying deep models on resource-constrained hardware; building datasets grounded in local contexts. Interested in the systems trade-offs between accuracy, latency, and power.
- Edge Inference
- Raspberry Pi
- Model Compression
ML Systems
Reproducibility, observability, and production-scale infrastructure that lets research ship. Informed by industry experience building distributed billing platforms with strong operational guarantees.
- Reproducibility
- Observability
- MLOps