Karthik Shivashankar

PhD Researcher

University of Oslo, Norway | 2021 – 2025

My research involved the end-to-end development of AI-driven tools to automate and enhance software quality analysis.

• Automated NLP-based Quality Analysis: Engineered BEACon-TD, a novel framework using fine-tuned transformer models to classify 13 distinct types of technical debt directly from unstructured text in issue trackers. This work was operationalized in the open-source TD-Suite framework.
• ML-Specific Code Quality & Maintainability: Addressed the lack of specialized quality tools for AI systems by building MLScent, a static analyzer that detects 76 unique ML-specific anti-patterns in frameworks like TensorFlow and PyTorch. This was complemented by a comprehensive systematic literature review to consolidate knowledge on maintainability in ML.
• Advanced Python Code Smell Detection: To overcome the limitations of existing linters, I developed PyExamine, a multi-level static analysis tool for Python that identifies 49 unique metrics covering code, structural, and architectural smells.
• Enhanced Code Maintainability with LLMs: Investigated the use of fine-tuned Large Language Models to proactively refactor Python code, demonstrating measurable improvements in objective maintainability metrics while preserving functionality.

AI Engineer

Fantastec Sports Technologies, London, UK | June 2018 - Oct 2021

• Automated digital asset creation by architecting a computer vision pipeline using Python, OpenCV, and Dlib, significantly reducing manual production time and costs.
• Constructed a scalable data pipeline on AWS with Apache Spark and Scikit-Learn to analyze user behavior, enabling a predictive in-app asset recommendation system.
• Designed a Monte Carlo simulation in MATLAB to forecast game dynamics for a blockchain application, delivering key data that guided product strategy and feature development.