About Me
I am Ravindu Tharuka Weerasinghe, an Artificial Intelligence undergraduate at the University of Moratuwa (CGPA: 3.86, Dean's List 4/6 semesters), with hands-on experience across small language models, MLOps, embedded systems, and production AI deployment.
My work is driven by a single question: what does it take for an AI system to actually work in the real world? That means thinking beyond benchmark accuracy — about latency, resource constraints, deployment pipelines, and the physical environments AI systems eventually have to operate in. I'm particularly drawn to the intersection of AI and robotics, and to making AI smaller, faster, and more explainable.
Experience
WSO2 — Research & AI Intern
Feb 2025 – Aug 2025
Worked within the AI research team on small language model fine-tuning, MLOps pipeline automation, and production AI tooling for the Choreo platform.
- Researched and applied fine-tuning techniques — LoRA, Adaptor, and Prefix Tuning — to domain-adapt language models to Choreo documentation; measured improvements using RAGAS evaluation metrics, achieving a lift from ~60% to ~65%
- Co-developed a fully automated MLOps pipeline: GitHub commits trigger RunPod GPU instance provisioning, run fine-tuning jobs, push trained models to Hugging Face Hub, and log all runs to a self-hosted MLflow server — end-to-end without manual intervention
- Designed and deployed Oxy's Gallery — a FastAPI image service hosted on WSO2 Choreo for WSO2Con Asia, accepting live images from a Unitree Go2 robot, displaying them in a responsive gallery, and enabling attendees to download images via QR code
- Developed the HubSpot video conferencing connector in Ballerina
- Focused throughout on privacy-sensitive and resource-constrained AI applications
Education
University of Moratuwa
B.Sc. (Hons) in Artificial Intelligence · 2022 – Present
CGPA: 3.86 / 4.00 · Dean's List: 4 of 6 semesters
Core areas: Neural Networks, Machine Learning, NLP, Computer Vision, Robotics, Software Engineering, and AI Systems.
BCS, The Chartered Institute for IT
Professional Graduate Diploma in IT · 2021 – 2025
Advanced Database Management Systems, Big Data Management, Programming Paradigms, and Network Information Systems.
SLIIT
Certificate in Cybersecurity & Digital Forensics · 2021 – 2022
Ananda College
GCE Advanced Level · Z-score: 2.37 · Combined Mathematics, Physics, ICT
GCE Ordinary Level · 9A passes
Research
Type II Diabetes Risk Prediction: A Multifactor Approach Using Laboratory and Lifestyle Features
IEEE Conference Publication · 2026
- Analysed 100,000+ NHANES records (1988–2018) for multiclass diabetes prediction: Not Diabetic, Type 2 Diabetes, and Other
- Designed a multi-stage feature selection pipeline (XGBoost, SHAP) reducing ~5,000 candidate features to 43 clinically practical predictors obtainable in real settings
- Achieved 95.27% ROC-AUC with 88.48% sensitivity and 87.19% specificity on the real-world held-out test distribution
- Built the system to handle missing data without imputation — a deliberate design choice for practical clinical deployment
- Applied Optuna hyperparameter search with 10-fold stratified cross-validation and full MLflow experiment tracking for reproducibility
Small Language Models for Privacy-Critical Applications
Independent Review · 2025
- In-depth review of SLMs for privacy-sensitive edge deployment — covering compression techniques including quantization, distillation, and pruning
- Analysed privacy-preserving frameworks: Differential Privacy, Secure Multi-Party Computation, Federated Learning, and Homomorphic Encryption applied to on-device inference
- Reviewed emerging directions including agentic SLMs and multi-agent systems operating at the edge
Co-curricular
AI Project Coordinator
Faculty of IT, University of Moratuwa · 2024 – Present
Coordinating AI-focused student projects and initiatives within the faculty.
IES Labs — R&D Pillar
Faculty of IT, University of Moratuwa · 2024 – Present
Batch 22 Mentorship
Faculty of IT, University of Moratuwa
Mentored junior students across two tracks — software projects (2024–2025) and hardware projects (2023–2024).
Competitions
Interests
- Efficient AI systems — small language models, quantization, on-device inference
- Robotics and embodied AI
- Neuro-symbolic reasoning and hybrid AI architectures
- Machine perception and computer vision
- Real-world deployment in resource-constrained environments
Personal Note
The project that shaped how I think about AI most is AutoChess — a fully autonomous chess-playing machine built and exhibited at EXMO 2023. I led the team as electronics designer, coded the sensing module, and worked through every layer from KiCad PCB design to embedded C++ firmware. Getting a physical system to work reliably — not just in a demo but consistently, across sixty-four squares and every edge case in chess — taught me things that purely software work simply doesn't.
That experience anchors a lot of what I find compelling now: AI that functions under real constraints, with real failure modes, in the real world. It's a harder problem than most benchmarks suggest.
Outside of work, I play badminton and guitar, and tend to read a lot of mystery novels. They share something with engineering — the satisfaction of a well-constructed problem, and a solution that feels inevitable once you see it.