Home icon

Ioanna Karagianni

Hello World! My name is Ioanna and I am from Athens, Greece. I completed my undergrad in Greek Philology with specialization in Linguistics at the University of Athens and currently I am in the last semester of my studies in Computational Linguistics with specialization in Speech Processing at the University of Stuttgart.

Decorative doodleI am keen about all topics related to NLP, especially NLU, as well as Speech Processing and more specifically TTS and ASR.

Decorative doodleI have also solid foundation and project experience in Machine Learning, Deep Learning and Data Analysis, topics that constitute the foundation of modern NLP applications.

Decorative doodleCurrently my work is more focused on the development of GenAI applications, RAG, Multi-Agent Systems and Agentic AI and how these can be utilized transparently and efficiently in enterprises.

Decorative doodle In my free time I enjoy traveling the world , reading books Decorative doodle, exploring new bookstores and coffee shops Decorative doodle, as well as watching movies or going to the cinemaDecorative doodle

Decorative doodle I also try to find time for pilates and yoga Decorative doodle and I am a firm believer of the importance of self care and nutrition Decorative doodle as they help me be productive and bring balance in my demanding schedule.

Feel free to reach out to me at joannakaragianni06 gmail.com and connect if you want to talk about anything related to the topics I mentioned, or just to say hi! Keep scrolling to learn more about me! Decorative doodle

Professional Experience

Decorative doodle Creation of Knowledge Base on AWS Bedrock and development of a RAG application for code generation on Software Defined Vehicles. Development of the automated generated code quality assessment metrics (SonarQube, CodeBertScore, CodeBleu).

Decorative doodle Created a questionnaire on Wix using JavaScript for the assesment of generated code quality by automotive experts based on ISO standards and conducted the interviews.

Decorative doodle Created and experimented with flows on PowerAutomate, connected to SharePoint, Teams & Outlook for the automation of different processes for colleagues.

Decorative doodle Conducted extensive research on Multi-Agent Systems and Agentic AI to detect commonly used patterns and developed a pipeline that parses through GitHub repositories that utilize MAS and Agentic AI, detects the use of patterns collected through the literature review process and stores them in a Database.

Decorative doodle Development of an enhanced RAG approach for assisting in the type approval process of SDVs by OEMs.

HiWi - University of Stuttgart Decorative doodle

August 2024 - September 2024

Participated in the annotation of data on a Procedural Learning Project under Dr. Karina Silberer.

Cabin Crew Member - Olympic Air Decorative doodle

March 2023 - October 2023

Fullfiled a long time dream of mine, to work as cabin crew member.

Decorative doodle Demonstration of safety procedures and emergency protocols.

Decorative doodle Committed to delivering exceptional experiences for passengers while ensuring their safety and comfort.

Decorative doodle Collaboration with flight crew, catering and operations team to optimise flight operations and improve quality of service.

Private Tutor Decorative doodle

October 2018 - June 2023

Tutored junior high school and high school students in Ancient Greek, Latin and Modern Greek Language and Literature.

Education

Leap of faith in tech, combining my great interest for technology and my love for language.

Courses of interest: Information Retrieval and Text Mining • Deep Learning for Speech and Language Processing • Advanced Speech Perception • Text Technology (HTML, XML, SQL, Relational Databases) • Computational Linguistics Team Laboratory (Text Emotion Detection) • Ethics and NLP • Interpretability and Analysis of NLP Models • Multilingual Speech Processing (Multilingual TTS) • Current Topics in Speech Processing (Immersive TTS) • Bayesian Statistics • Foundational Questions regarding LLMs • AI Prototyping, Technology Entrepreneurship, Project Management

6 - month R&D Project under Dr. Vu, Thang: Fine-Grained Controllable LLM based Speech Synthesis via Instruction Tuning.

Aim of the project: To investigate if fine grained controllability in LLM-based TTS systems is possible. We did so by combining CosyVoice2 with the ParaCLAP framework. The goal was to control speech characteristics such as pitch, duration, intensity, jitter, and shimmer using natural language instructions at both utterance and word level. Our approach leveraged acoustic feature extraction and instruction tuning to enable more expressive and controllable speech synthesis.

Thesis: Beyond EER: Speaker-Level Evaluation and Performance Assessment of Voice Anonymization with Alternative Metrics and Phonetic Features Analysis.

Thesis Scope: Examination of alternative evaluation metrics for assessing privacy in voice anonymization systems in order to overcome the shortcommings of EER. The aims to provide a deeper understanding of how these metrics capture privacy preservation compared to the traditional EER used in the VPC 2024. Moreover, an effort is made to investigate the meta-level, primarily the phonetic but also the potential linguistic factors at the speaker level that may affect the effectiveness of anonymization, either positively or negatively.

University of Athens - BA. Greek Philology Decorative doodle

October 2017 – September 2021

GPA: 8.32/10 (top 10% of the department)

Courses of interest: Introduction to Linguistics • Historical Linguistics, Introduction to IE Linguistics • Phonetics- Phonology (lab) • Morphology • Syntax I • Semantics • Pragmatics •Historical Grammar of Ancient Greek (Phonetics-Phonology) • Psycholinguistics- Neurolinguistics • Language Acquisition • Computational Linguistics • Second Language Acquisition - Teaching Greek as a Foreign Language • History of Linguistics • Text Linguistics- Discourse Analysis • Experimental Linguistics • Sociolinguistics

Thesis: The Argument Structure production in the narrative speech of native Greek patients with Broca’s and Anomic Aphasia in association with Athena Research Centre

Societies: Lost Student´s Society - Lingustics Journal Club, Accessibility Unit.

One of the 30 finalists selected from 500 applicants to participate in the intensive bootcamp by WE LEAD, in collaboration with Code.Hub.

Topics Covered: Database Management, Data Visualisation, ML Libraries, Soft skills sessions offered by Morphoses on Critical Thinking, Teamwork and Negotiation.

Final Group Project: Predicting the City-Cycle fuel consumption in miles per gallon of a car - A Classification Problem, where I implemented a NN model using torch that achieved an F1 score of 0.98.

Projects

The Role of Milk Energy in the Development of Primates - Bayesian Statistics Course Decorative doodle

February 2025 - March 2025

The goal of this experiment was to model the kilocalories of energy in milk from neocortex mass and average female body mass on a dataset based on female primate data from McElreath (2020). In this project I performed the EDA, the Bayesian imputation and multivariate Bayesian Regression using PyMC.

Predicting the City Cycle Fuel Consumption in MPG of a Car - A Classification Problem Decorative doodle

December 2024 - February 2025

Implemented a Multi Layer Perceptron using torch, Cross Entropy Loss and Adam Optimizer. To avoid overfitting I used the ReduceLROnPlateau Scheduler and Early Stopping. The model achieved an F1 score of 0.98, the highest among all 5 groups.

Intent Clustering for Telecommunication Company - Hackathon by Netcompany Decorative doodle

January 2025 - January 2025

Project developed for a Hackathon Event offered by Netcompany on the Bootcamp: Data Science and Business Intelligence offered by WE LEAD & Code.Hub The goal was to create Clustering-Based Intent Identification Engine for Greek, which would cluster a demand placed by a customer. Developed a Clustering Algorithm using K-means. Utilized Multilingual SBERT for embeddings, UMAP for Dimensionality Reduction & Silhouette Score for finding optimal number of clusters and Evaluation.

Geognossis: Leveraging XML & MongoDB for Real-Time Data Retrieval - Text Technology Course Decorative doodle

May 2024 - August 2024

A tourism attraction finder that uses the Overpass API and aims to provide the user the opportunity to find the tourist attractions in an area by name or coordinates. We used Flask for building the backend, stored and cached the attraction data in MongoDB and parsed the data retrieved from the Overpass API in XML format using ElementTree. The frontend was built using Leaflet.js for interactive maps and Geopy for geocoding user inputs.

Emotion Detection on Text, Multi-label Classification - Computational Linguistics Lab Decorative doodle

April 2024 - August 2024

From scratch implementation of a Baseline Classifiers and Evaluation Methods: Binary & Multi-label perceptron and Precision, Recall, F1-Score. Then I developed more advanced approaches for the classification using LSTM and Bi-LSTM & use of overfitting prevention techniques (Early Stopping, Dropout) & comparison of model performance with different types of embeddings: TF-IDF, GloVE, BERT. For these imlementations I used mainly the libraries transformers, tensorflow and sklearn.

Speech Emotion Recognition - Deep Learning Course Decorative doodle

February 2024 - March 2024

Classification problem where we needed to predict the emotion class based on speech input features. There are 4 classes where a two-dimensional representation was used. I used a Bi-LSTM model for the emotion classification. The model is trained to predict valence and activation states from the given features. The dataset used includes training, validation, and testing splits in JSON format. The libraries I mainly used for implementation and evaluation were torch and sklearn.

Design & Implementation of N-Gram model in Python - Programming in Python Course Decorative doodle

February 2024 - March 2024

I built a program that can learn statistical patterns from text and use those patterns to predict the likelihood of certain words following others, using the library NLTK.

Publications

Structured Knowledge for Complex Domains: A Generative AI Pipeline for Software-Defined Vehicle Application Development Decorative doodle

IEEE Smart Mobility 2026

(Accepted) Cheng, X., Karagianni, I., Morar, D., and Slama, D., Structured Knowledge for Complex Domains: A Generative AI Pipeline for Software-Defined Vehicle Application Development, Track 1: Smart Mobility Technologies, 2026

Probing Discrete Speech Tokens of Spoken Language Models Decorative doodle

LREC 2026

(Accepted) Naber, S., Koch, J., Singh, P., Saponaro, A., Karagianni, I. and Vu, T. Probing Discrete Speech Tokens of Spoken Language Models, Language Resources and Evaluation Conference (LREC), 2026

Volunteering

Volunteer in the Speakers Team of the WiDS Zurich Chapter, responsible for identifying and reaching out to potential speakers for the annual WiDS Zurich Conference, as well as for helping with the organization of the speaker sessions during the conference.

Outgoing Global Volunteer - AIESEC in Germany Decorative doodle

November 2023 - October 2024

Assisting applicants in the application process for the Global Volunteer program, providing them with all the necessary information and support to ensure a smooth application and follow up process.

PR Manager - LingUU Journal Decorative doodle

May 2022 - August 2023

Kept the journal's WordPress site up to date and I was also responsible for the public image of the journal by updating its social media platforms.

Elected Manager at the Focus Area of Education, coordinating team members in order to promote educational content. Created YouTube videos and online live talks by junior researchers and postgraduate students presenting key points of their research publications or scientific papers. I was also responsible for the script writing for YouTube videos, including "The case of Phineas Gage" and "What is consciousness".