Madusha Sandaruwan - Portfolio Website

What I Work On

Bridging the gap between backend systems and user experiences.

System Architecture

Designing distributed systems and event-driven architectures focused on reliability, maintainability, and observability for high-throughput services.

Kafka Redis gRPC

AI / ML & Research

End-to-end ML systems: data preprocessing, modeling, experiment tracking, evaluation, and production deployments. Research experience with CV+NLP fusion models and conference publications.

PyTorch Transformers TensorFlow

Backend Engineering

Developing high-performance APIs, optimizing data access, and implementing robust service patterns for resilience and scale.

C# PostgreSQL GraphQL

DevOps, Infra & Security

Implementing cloud infrastructure (IaC), CI/CD pipelines with automated testing and supply-chain security to reduce release risk and improve reliability.

Azure Docker Terraform

Featured Case Study

Solving complex problems with research-driven methodology.

Research 2025 // Healthcare

Multi-modal Alzheimer’s Classification with Cross-Attention Fusion

Role: Lead Researcher

The Challenge

Alzheimer’s Disease (AD) detection is challenging in early stages due to heterogeneous symptomatic progression. Standard methods often rely on single modalities, failing to capture the complex interplay between neuroimaging markers and clinical variables.

The Solution

Developed an Enhanced Multi-Modal Pipeline utilizing the OASIS-3 dataset. I implemented a structural alignment process to the 1mm MNI152 atlas, extracting 10 coronal slices at specific Y-coordinates optimized for hippocampal and cortical atrophy detection.

To handle sparse clinical data, I designed a longitudinal-aware imputation strategy. The architecture employs a dual-stream encoder: 10-channel EfficientNet-B0 for MRIs and Bio-Clinical-BERT for structured clinical narratives, integrated via a Cross-Attention Fusion hub.

The Outcome

→ Achieved average AUC of 0.8952 and accuracy of 87.35% (5-fold Group K-Fold).
→ Validated against longitudinal subject trajectories (MCI to AD converters).
→ Accepted for publication at ICATC 2025 (IEEE Xplore).

PyTorch Transformers EfficientNet-B0 Bio_ClinicalBERT MNI152-Atlas Scikit-learn

5-Fold Group K-Fold (Multi-Modal)

T1w MRI (10 Slices)

Clinical Data

MRI Clean Up & Process

Skull Strip & 2D Extraction

Narrative Gen

Clinical Text Synthesis

EfficientNet-B0 (Image Encoder)

Bio_Clinical_BERT (Text Transformer)

Cross Attention Fusion

Multi-Modal Integration

Majority Vote
Ensemble

Average
Ensemble

About Me

The human behind the code.

[Profile_IMG]

Location Colombo, Sri Lanka

Status AI and Software Architect

Focus Computer Vision & NLP Research

"Good software is boring. It just works, and nobody notices."

01 // Biography

I'm a founding engineer at Wirity Labs. My days are full of writing code, designing systems, and figuring out elegant solutions for complex problems. This all started back in 2013 when curiosity got the best of me, I wanted to understand how things worked at a deeper level. These days, my focus is on backend architecture, secure coding, and infrastructure. I enjoy working on problems that ensure systems run reliably over time.

When I'm not coding, you'll find me diving into stuff about history or philosophy. It's interesting how understanding both technical and human systems benefits from learning to see the bigger picture.

Years Experience 2+

Projects Delivered 10+

Certifications

Publications 1

02 // Experience

Wirity Labs

2025 – Present

Founding Engineer

Architected secure backend infrastructures with a focus on system reliability and maintainability.
Managed large-scale software repositories, monorepos and version control systems, ensuring code quality through automated pipelines.

Hono .NET 9 Typescript Cloudflare PostgreSQL React Docker

Perituza Software Solutions

2024 – 2025

Information Technology Intern (.NET/ R&D)

Built a .NET benchmarking suite with BenchmarkDotNet to compare EF Core and ADO.NET on MySQL, PostgreSQL, and SQL Server, guiding future project options.
Automated CI/CD in GitHub Actions with unit tests, static analysis, security scans, and Jira and MS Teams notifications, cutting manual steps and reducing release issues.

C# .NET Azure DevOps GitHub Actions Docker SQL