Research Profile
I am an AI Researcher specializing in the convergence of Machine Learning (ML), Natural Language Processing (NLP), and Software Engineering (SE). My doctoral research focuses on managing Technical Debt in modern software systems by leveraging Transformers and Large Language Models (LLMs). I have designed and deployed agentic workflows and static analysis tools (BEACon-TD, MLScent) that automate the identification of code smells and architectural debt. My expertise spans the full lifecycle of AI engineering, from training specialized language models to developing scalable detection frameworks for ML-specific anti-patterns.
Education
Philosophiae Doctor (PhD) in Computer Science
Thesis: The Dual Role of Machine Learning in Technical Debt Management.
Developed specialized BERT-based models for classifying technical debt and created a taxonomy of 76 ML-specific anti-patterns.
MSc in Electronics Engineering (Distinction)
Dissertation: Deep Learning for 4D Video.
Developed compact representations of 4D video sequences using 3D Variational Autoencoders (VAEs) for Mixed Reality rendering.
Research & Technical Experience
PhD Researcher (AI & Software Engineering)
2021 – 2025University of Oslo
- Transformer-Based Debt Classification: Engineered BEACon-TD, a novel framework utilizing fine-tuned transformer models to classify 13 distinct types of technical debt from issue trackers.
- Agentic AI for Refactoring: Investigated Agentic AI workflows using fine-tuned LLMs to proactively identify and refactor Python code, demonstrating measurable improvements in Halstead metrics.
- ML Static Analysis: Architected MLScent, a static analysis tool designed to detect 76 unique ML-specific anti-patterns in frameworks like TensorFlow and PyTorch.
- Python Quality Assurance: Developed PyExamine, a multi-level static analysis tool identifying 49 metrics for code and architectural smells, achieving 91% recall.
AI Engineer
2018 – 2021Fantastec Sports Technologies
- Architected a computer vision pipeline using Python, OpenCV, and Dlib that automated digital asset creation, significantly reducing manual production overhead.
- Constructed scalable data pipelines on AWS and Databricks using Apache Spark to analyze user behavior for predictive asset recommendation.
Technical Skills
Core AI & ML
Software Engineering
Stack
Selected Publications
BEACon-TD: Classifying Technical Debt using transformers
K. Shivashankar, M. Orucevic, et al. • Journal of Systems and Software, 2025