Project detail
Generating Distributed Code from Single Node Code
Final Year Project Showcase poster on graph-based monolith-to-microservices decomposition using structural and semantic program signals.
Role
Research contributor and presenter
Status
Presented at FYP Showcase 2026
Focus
Software architecture, graph learning, and microservice decomposition
Overview
Presented at the Final Year Project Showcase 2026, this research poster introduces an automated graph-based framework for identifying microservice boundaries from monolithic systems. The approach models program entities, resource entities, CALLS relations, and CRUD relations as a heterogeneous graph, learns embeddings with an edge-aware graph autoencoder, preserves semantic similarity with triplet loss, and clusters the learned representations to generate candidate microservices.
Outcome
A research-backed decomposition framework that produced stronger composite scores on benchmark systems compared with selected baseline approaches.
Highlights
Integrated structural, semantic, and domain-related program signals into a unified heterogeneous graph representation.
Used an edge-aware graph autoencoder to learn compact embeddings for program and resource entities.
Evaluated decomposition quality with SM, ICP, IFN, NED, and a weighted composite score across benchmark systems.
Presented the work with Mineth Weerasinghe, Methmini Madhushika, and Dhanuka Lakshan at the University of Moratuwa FYP Showcase.
More projects