Software Developer @ SimHealthAI · Computer Engineering, University of Jaffna · Sri Lanka 🇱🇰

I build systems where machine learning has to survive contact with real constraints — privacy regulation, distributed trust, and production infrastructure.

My work sits at the intersection of applied ML and decentralised systems: federated learning across hospitals, blockchain-verified credentials, and embedded IoT telemetry.

Languages

ML / Data

Backend & Infra

Distributed & Web3

Frontend

Hospitals collaboratively train a COVID-19 chest X-ray classifier without sharing patient data. Local training on non-IID data → (ε,δ)-differential privacy applied to weight deltas → IPFS storage → every model CID recorded immutably on Hyperledger Fabric.

Implemented FedProx against a FedAvg baseline to counter client drift on heterogeneous hospital data — ~4.5% higher accuracy at convergence. Ships with a live dashboard for convergence and privacy-budget tracking.

Python PyTorch Hyperledger Fabric IPFS Node.js Docker

Universities mint degree certificates as Cardano NFTs; employers verify them instantly against on-chain metadata. Wallet integration via Mesh SDK (Eternl, Lace), verification through the Blockfrost API.

Built in 2 weeks with a team of 5 for UniHack 2025, the inter-university blockchain hackathon (CoinCeylon × University of Kelaniya).

React TypeScript Cardano Mesh SDK Tailwind

Encoder–decoder model that describes images in natural language. Pretrained InceptionV3 extracts 2048-d visual features, merged with tokenised text and decoded word-by-word by an LSTM. Trained on Flickr8k (8,091 images × 5 captions).

Python TensorFlow Keras Computer Vision NLP

ESP32-based smart dustbin using ultrasonic fill-level sensing, GPS, and SIM800L GSM to send SMS alerts when bins are full — cutting unnecessary collection trips.

C ESP32 Embedded IoT