Petros Mitseas
I build AI systems that work outside the lab.
I'm a machine learning engineer in Athens. I've spent my last few years building AI for emotion and speech recognition, healthcare and agentic workflows, along with the software that takes it from research to production.
Timeline
What I've been building
2025 – now
AI Lead · Aisthesis Medical
I lead the development of a model that predicts sepsis, a life-threatening reaction to infection, from real hospital data, and I set up how the team works together day to day.
Architected and co-developed the full pipeline: data extraction, preprocessing, feature engineering, deep learning training, evaluation and reporting. Set up team infrastructure and engineering workflows, and contributed to integrating the model into production.
2025
Pharos · Break-Even Consulting
I designed and built Pharos, an assistant that finds the right information across a company's documents. It went live for two enterprise customers within months.
Architecture and development of an LLM-based retrieval platform (RAG) over enterprise knowledge bases of up to a hundred thousand documents.
2022 – 2025
Machine Learning Engineer, later Senior · Behavioral Signals
I worked on AI that recognises emotions and behaviour from the sound of a person's voice, and on detecting deepfakes, and helped take both from research into products that customers use.
Research, development and deployment of state-of-the-art deep learning audio models for emotion and behaviour classification. Built annotation, data and training infrastructure, designed low-latency real-time inference APIs and led the AI-Labs initiative. Published at INTERSPEECH and ACII.
2019 – 2022
Software Engineer · Cognity
I built the systems behind telecom customer portals, where orders are placed, customers are managed and bills are issued, and helped move old, monolithic software to a modern, modular design.
Java, Spring Boot and Spring Cloud in large, multi-team Scrum projects. Migrated legacy monoliths to microservices and set up CI/CD pipelines.
2017
Embedded Systems Engineer · Kenotom
My first engineering job: writing the software that runs on small chips inside connected devices, and designing how those devices talk to each other.
C on ARM microcontrollers (development, debugging and unit testing), and the communication infrastructure for a network of connected microcontrollers.
Education
Where I studied
2021 – 2023
MSc in Artificial Intelligence · NCSR "Demokritos" & University of Piraeus
Alongside my job, I went back to study AI. For my thesis, I taught a computer to recognise when two recordings are different versions of the same song.
Thesis on deep learning methods for cover song identification.
2012 – 2018
Diploma in Electrical & Computer Engineering · Aristotle University of Thessaloniki
My engineering degree. For my thesis, I taught a robot to grasp objects by learning from trial and error.
Integrated MEng. Thesis on applying reinforcement learning to robot grasping.
What am I looking at?Picture every way an AI could do a task as a spot on this landscape, where the height shows how many mistakes it makes. Learning is like a ball rolling downhill: step by step it finds the lowest valley, where it gets the most answers right.
Research
Published papers
2024Best Paper Award
Emotion-Aware Speech Popularity Prediction: A Use-Case on TED Talks
Predicting how popular a TED talk will be from the emotions in the speaker's voice.
12th International Conference on Affective Computing and Intelligent Interaction (ACII), Glasgow, pp. 300–308
2024
RobuSER: A Robustness Benchmark for Speech Emotion Recognition
A benchmark that tests how reliably emotion-recognition models cope with imperfect, real-world audio.
12th International Conference on Affective Computing and Intelligent Interaction (ACII), Glasgow, pp. 219–227
2023
Cross-Lingual Features for Alzheimer's Dementia Detection from Speech
Detecting signs of Alzheimer's dementia in the way people speak, using features that work across languages.
Proc. Interspeech 2023, pp. 3008–3012