Research Scientist · Oslo

Karthik Shivashankar

I work at the intersection of Generative AI, Agentic Engineering, and Software Engineering and Quality.

● SINTEF Digital ● PhD, University of Oslo

01

Research focus

// what I study

Generative AI Agentic Engineering LLMs Transformer models Technical debt classification ML maintainability & scalability Software quality Code smells & anti-patterns Static analysis Issue-tracker analytics Task prioritisation Natural language processing Deep learning Open-source tooling

02

Experience

// where I build

2025 — Present
Current

Research Scientist

SINTEF Digital — Sustainable Communication Technologies — GIoT Group

Applied research on generative and agentic AI for software engineering.

2021 — 2025

PhD Researcher — AI & Software Engineering

University of Oslo, Department of Informatics

Thesis: The Dual Role of Machine Learning in Technical Debt Management.

BEACon-TD — fine-tuned transformer framework classifying 13 distinct types of technical debt from issue trackers.
Agentic refactoring — LLM-driven workflows that proactively refactor Python, with measurable gains.
MLScent — static analysis detecting 76 ML-specific anti-patterns across TensorFlow and PyTorch.
PyExamine — multi-level smell detection across 49 metrics, reaching 91% recall.

2018 — 2021

AI Engineer

Fantastec Sports Technologies

Built a computer-vision pipeline (Python · OpenCV · Dlib) automating digital-asset creation and cutting manual overhead.
Constructed scalable data pipelines on AWS & Databricks with Apache Spark for predictive Bloackchain asset recommendation.

03

Education

// where I learned

2021 — 2025

PhD in Computer Science

University of Oslo · Department of Informatics

Developed specialised BERT-based models for classifying technical debt and a taxonomy of 76 ML-specific anti-patterns.

2017 — 2019

MSc Electronics Engineering — Distinction

University of Surrey

Dissertation: Deep Learning for 4D Video — compact 4D sequence representations via 3D Variational Autoencoders for Mixed-Reality rendering.

04

Technical toolkit

// how I build

Methods

LLMsAgentic AITransformers NLPComputer VisionDeep Learning Static AnalysisTechnical DebtCode Smells

Stack

PythonPyTorchHugging Face Scikit-learnSQLDocker AWSSparkMLOps · CI/CD

05

Selected publications

// full list on Scholar →

TD-Suite: An All-Batteries-Included Framework for Technical Debt Classification

K. Shivashankar & A. Martini · Preprint / framework release

2025

Maintainability & Scalability in Machine Learning: Challenges and Solutions

K. Shivashankar, G. S. Al Hajj & A. Martini · ACM Computing Surveys

PyExamine: A Comprehensive, Un-Opinionated Smell Detection Tool for Python

K. Shivashankar & A. Martini · Proceedings of MSR

MLScent: A Tool for Anti-Pattern Detection in ML Projects

K. Shivashankar & A. Martini · Proceedings of CAIN

2025

BEACon-TD: Classifying Technical Debt and its Types Across Diverse Projects using Transformers

K. Shivashankar, M. Orucevic, M. M. Kruke & A. Martini · Journal of Systems and Software

2025

Enhancing Python Code Maintainability through Large Language Model-Based Approaches

K. Shivashankar & A. Martini

2025

Enhancing Task Prioritization in Software Development Issue-Tracking Systems

K. Shivashankar, K. M. Haugerud & A. Martini

2025

QualiTagger: Automating Software-Quality Detection in Issue Trackers

K. Shivashankar, R. Capilla, M. M. Kruke et al.

2023

Technical Debt Classification in Issue Trackers using Transformer-Based NLP

K. Shivashankar & A. Martini

2022

Maintainability Challenges in ML: A Systematic Literature Review

K. Shivashankar & A. Martini

06

Open-source

// ship it

TD-Suite

Transformer-based technical debt classification across 17 categories — batteries included.

text_classification

Static Analysis

PyExamine

Comprehensive, un-opinionated code-smell detection for Python across 49 metrics.

python_smells_detector

MLScent

Detects 76 ML-specific anti-patterns across TensorFlow and PyTorch codebases.

ml_smells_detector