Curriculum Vitae
Research, Publications & Experience
Andreas Avgoustis
PhD Candidate — Visual Speech Recognition & Multimodal Learning
RESEARCH PROFILE
PhD candidate working on visual speech recognition for people who have lost their voice after tracheostomy or laryngectomy. My research asks whether a vision–language model, rather than a conventional video classifier, can recognise Greek speech from lip movement alone with enough accuracy to be usable at the bedside. I built the first Greek viseme-level dataset for this population, designed a deterministic preprocessing pipeline that encodes a video segment as a single image a VLM can ingest, and combined parameter-efficient fine-tuning with dense retrieval and a Greek language model as a correction layer. My interests are multimodal representation learning, efficient adaptation of large pretrained models, and evaluation of these systems under the atypical articulation that real patients present.
BiHELab, Ionian University
Supervisor: Prof. Themis Exarchos
Greek (Native), English (C2)
Research
May 2024 - Present
Doctoral Researcher — Visual Speech Recognition
BiHELab (Bioinformatics & Human Electrophysiology Laboratory), Ionian University
- Developing a personalized Greek lip-reading system for patients who lose the ability to speak after tracheostomy, from data collection through to evaluation.
- Built the Greek word-level video corpora the work relies on multiple speakers, several hundred distinct words each together with an automated preprocessing pipeline for mouth-region extraction, temporal resampling and recovery of frames where face detection fails.
- Designed a viseme-level recognition model a self-supervised vision encoder,a bidirectional recurrent sequence model and a CTC objective that composes words from visual units instead of matching whole recordings, so vocabulary never recorded in training remains reachable: 11.4% top-1 and 20.3% top-5 on unseen words against a chance level of 0.003%.
- Implemented and benchmarked an alternative retrieval architecture (vision–language model with 4-bit QLoRA, FAISS dense retrieval and a Greek language model as a correction layer) across four speakers, and used the comparison to establish which approach the thesis should pursue.
- Designed the evaluation protocol: open-vocabulary tests against permutation baselines, multi-seed runs with reported variance, and per-speaker analysis of failure cases.
Publications
- Visual Speech Recognition for Enhancing Communication in Post-Tracheostomy PatientsAccepted · in press Avgoustis, A.; Pezoulas, V.; Exarchos, T. GeNeDis 2026 — World Congress on Genetics, Geriatrics and Neurodegenerative Diseases Research
- Bridging Linguistic Gaps: Developing a Greek Text Simplification Dataset Agathos, L.; Avgoustis, A.; Kryelesi, X.; Makridou, A.; Tzanis, I.; Mouratidis, D.; Kermanidis, K.L.; Kanavos, A. Information, 15(8), 500 (2024) doi.org/10.3390/info15080500
- Optimization of Healthcare Process Management Using Machine Learning Avgoustis, A.; Exarchos, T.; Vrahatis, A.G.; Vlamos, P. Artificial Intelligence Applications and Innovations (AIAI 2024), IFIP AICT vol. 711, Springer doi.org/10.1007/978-3-031-63211-2_15
- Identifying Earthquakes in Low-Cost Sensor Signals Contaminated with Vehicular Noise Agathos, L.; Avgoustis, A.; Avgoustis, N.; Vlachos, I.; Karydis, I.; Avlonitis, M. Applied Sciences, 13(19), 10884 (2023) doi.org/10.3390/app131910884
- Applied Deep Learning for Categorizing Dermoscopic Images Avgoustis, A.; Exarchos, T.; Kermanidis, K.L.; Mylonas, P. IEEE SMAP (2021), pp. 1–4 doi.org/10.1109/SMAP53521.2021.9610798
Technical Skills
Vision & multimodal
Vision–language models (LLaVA, BakLLaVA), QLoRA / parameter-efficient fine-tuning, CNNs, 3D-CNN–LSTM, Transformers, MediaPipe FaceLandmarker, OpenCV.
Language & retrieval
Retrieval-augmented generation, FAISS dense retrieval, sentence embeddings, Greek LLMs (Krikri), Hugging Face Transformers, PEFT, bitsandbytes.
Frameworks
PyTorch, TensorFlow, Keras, scikit-learn, NumPy, pandas, Apache Spark.
Infrastructure
Python, SQL, Git, Docker, cloud GPU training (Google Cloud Platform, Microsoft Azure, Colab), Linux.
Also
R, PHP, PowerShell, Power BI, Tableau, React, Django.
Additional Information
Conference presentations
Oral presentations at SMAP 2021 (online) and AIAI 2024 (Corfu, Greece).
Teaching support
Exam invigilation duties, Department of Informatics, Ionian University (2025–2026).
Certifications
Data Engineering, Full-Stack Python Developer, and Business Analyst Career Track — IT & Engineering School.
Professional Experience
April 2026 - June 2026
Adult Education Instructor
APOPSI IT, Consulting & Training Services S.A.
- Delivered 140 hours of in-person training for the national programme "Digital education and empowerment of older people and persons with disabilities" (MIS 5223341), at the Open Care Centre for the Elderly (KAPI) of Central Corfu and the Diapontia Islands.
- Taught digital skills to senior citizens: device use, safe internet navigation, online communication, e-government services and online banking.
Dec 2024 - Nov 2025
Computer Programmer — Compulsory Military Service
Research and Informatics Office, 8th Motorized Infantry Brigade, Hellenic Army
- Administered intranet servers, virtualised environments (VMware ESXi) and network infrastructure, with responsibility for availability, backup and disaster recovery.
- Automated recurring administrative workflows through PowerShell and Java scripting. Served concurrently with ongoing doctoral research.
Feb 2023 - Feb 2024
Business Intelligence Analyst
Association of Educational Caregivers of Corfu
- Built reporting pipelines and dashboards over web and social engagement data, and designed the organisation's data storage and update process.
Jan 2021 - Jan 2023
Junior Data Analyst
Alfa Ekpaideutiki
- Analysed web performance data with SQL and Excel, implemented PHP-based tracking with the development team, and produced recurring analytical reports for management.
Education
May 2024 - Present
PhD in Informatics (in progress)
Ionian University, Department of Informatics, Corfu
- Thesis: Models and Applications for Lip Reading and Voice Reproduction.
- Supervisor: Prof. Themis Exarchos. Thesis language: English.
Oct 2022 - Feb 2024
MSc in Big Data Driven Networked Systems
Ionian University, Department of Informatics, Corfu
- Grade 9.47 / 10. Thesis on machine learning for healthcare process optimisation, published at AIAI 2024.
- Coursework: big data technologies, advanced data mining and machine learning, stochastic analysis, network systems and security.
Oct 2015 - Feb 2021
BSc in Informatics
Ionian University, Department of Informatics, Corfu
- Thesis on deep learning for melanoma classification from dermoscopic images, published at SMAP 2021.