Available for select projects
Designing AI governance workflows and intelligent agents at Always Online.
01 · ABOUT
I study how teams work, then rebuild their workflows with AI.
I’m an AI Solution Architect working at the intersection of engineering and business strategy. I studied Management Science & Technology at AUEB, specialising in software engineering and data science, a combination that shapes how I approach problems.
My work starts with how teams actually operate today: the messy reality of decisions, handoffs, and judgment calls that don’t show up in process diagrams. From there, I redesign workflows around AI, not to replace people but to reshape what they spend their time on.
I currently lead AI solution design at Always Online, reporting directly to client executive teams. Previously at EY, where I built agentic systems and evaluation frameworks for banking and manufacturing clients.
Outside of consulting, I coauthored a paper on how AI agents reshape data scientists’ roles, exploring the same question that drives my client work, just in a more academic register.
Based between Athens and Patras.
02 · EXPERIENCE
Leading the design and project management of AI solutions and intelligent agents for enterprise clients. Reporting directly to chief executives of customer organisations.
Designing complete AI solutions for executive teams across multiple industries.
Contributed to technology consulting projects for banking and manufacturing clients. Built and evaluated internal AI solutions, presenting outcomes to EY’s AI department.
Built and evaluated agentic systems for banking and manufacturing clients.
Volunteered with INSEAD’s MBA course “Building Generative AI Products and Businesses”. Led the business and technical design of an MVP, working with MBA students to bridge product strategy and AI implementation.
Led business and technical design of an AI MVP with MBA students.
03 · EDUCATION & CERTIFICATES
Specialised in Software Engineering and Data Science within the Management Science & Technology department, a program that blends technical training with business strategy, operations, and analytics.
Research collaboration with Prof. George Lekakos on AI agents’ impact on data science roles, leading to a coauthored conference paper.
Certificates
Coursework in applying data analytics to business decision frameworks.
Certified in Salesforce CRM administration with focus on data management.
Coursework in negotiation theory and applied strategy.
04 · RESEARCH
EuroMed Academy of Business Conference 2026 · Paphos, Cyprus · 23 to 25 September 2026
Existing skills frameworks for data scientists were written before AI agents and still sort competence into technical and soft skills, so they miss how fast the role is actually changing. This paper examines six AI agent tools across the full data science pipeline to trace whether new tasks emerge along the way, and to point to the skills the role now requires. The analysis surfaces two emerging skills for data professionals: Agent Steering and Agent Process Tracing.
Data scientists are not replaced by AI agents; the centre of gravity of the role shifts from executing analysis to governing and validating autonomous execution.
Presented at the EuroMed Academy of Business Conference 2026, Paphos, Cyprus, 23 to 25 September 2026.
Data scientists are not replaced by AI agents; the centre of gravity of the role shifts from executing analysis to governing and validating autonomous execution.
05 · PROJECTS
Desktop application that assigns Erasmus students to partner universities from their academic ranking and their three ranked preferences. Delivered to my university's Erasmus office for use, replacing a placement they had been doing by hand in spreadsheets.
Two methods, side by side. The first automates the office's existing manual process as a priority heuristic, turning about a week of spreadsheet work into five minutes. The second, developed with an optimisation professor as part of his course, treats the whole cohort as an Integer Linear Program (Pyomo, GLPK solver) that maximises total student satisfaction subject to university capacities and pairs of students who must be placed together.
How far has Airbnb actually spread through Europe's cities, and do the numbers people quote about it hold up? Twelve months of listing data across thirteen cities, used to answer both.
Python and pandas over four seasonal snapshots per city. The central metric is density, listings per 1,000 residents, which meant matching each city's population to the exact map boundary its data covers. Listings that are genuinely booked are separated from dormant ones, price outliers are trimmed, and every booking estimate is cross-checked against an independent commercial source. Visualised with matplotlib and an interactive Vega-Altair dashboard.
Tool that builds the best value Euroleague fantasy roster from historical player statistics.
It is formulated as a binary integer programming problem in PuLP (CBC solver) that maximises weighted fantasy points under budget and position constraints. Because it is solved exactly, the roster it returns is provably optimal rather than a good guess.
Machine learning project predicting whether a breast mass is benign or malignant, on the UCI Mammographic Masses dataset.
After cleaning the data (handling missing values and selecting features), it benchmarks six classifiers, from Decision Tree, SVM, KNN, Naive Bayes and Logistic Regression to a Keras neural network, on accuracy, precision, recall and F1. The neural network came out on top at roughly 83% accuracy.
Open source AI terminal tool that helps engineers find jobs, using a large language model to rank and match postings against a profile.
My contribution: a web scraper for Y Combinator (Work at a Startup) postings on top of the existing Hacker News source. It greatly enlarged the job database and roughly doubled the tool's throughput.
My solution for the SemEval 2025 shared task on detecting food hazards and products from short food incident report texts.
I trained a large RoBERTa model (Hugging Face Transformers with PyTorch), one per subtask, using text augmentation and Focal Loss for class imbalance and mixed precision AdamW training. It reached a macro F1 of about 0.79 on the main subtask.
06 · CONTACT