Hi, I'm Achilleas Leivadiotis

A passionate Data Science & AI graduate specializing in machine learning, computer vision, AI engineering, and full-stack development.

View My Work About Me Contact Me

Machine Learning

code-editor

Python

1# AI model training

2model = Sequential([

3 Dense(128, activation='relu'),

4 Dropout(0.2),

5 Dense(10, activation='softmax')

6])

My Portfolio

From enterprise platforms to AI systems - explore my diverse project portfolio

Full-StackProduction Ready

Kassandra Properties VIP - Web App

Enterprise-grade construction & real estate platform featuring property listings, property management, and full Admin CMS with Supabase. Built with Next.js 15, React 19, Prisma, and Supabase.

Next.jsTypeScriptSupabasePrismaAdmin CMSNextAuth

View Live Site GitHub

Real Estate Platform

Bachelor ThesisGrade: 9/10

AI-Powered Aviation Contrail Monitoring System

EUROCONTROL MUAC • Feb 2025 - July 2025

40%

of aviation warming

from contrail-cirrus

Breakthrough Achievement

Engineered the first validated end-to-end AI pipeline for ground-based contrail monitoring, combining computer vision, multi-object tracking, and flight data correlation. This pioneering prototype transforms ground-camera imagery into flight-resolved contrail datasets with 71.75% attribution accuracy, a major leap beyond prior work. The system enables real-time monitoring and mitigation strategies for aviation's hidden climate footprint, where contrails contribute ~40% of aviation's total warming effect.

Download Full Thesis (PDF)

Multi-Object Tracking Demonstration

Norfair tracking algorithm demonstrating superior performance with MOTA: 43.3% and IDF1: 67.9%

Stage 1: Detection

Custom Detectron2 Pipeline

✓

Mask R-CNN (ResNet-101 FPN) optimized for contrail morphology

✓

~65% mask mAP for segmentation

✓

>90% class accuracy for short, thin, wide, cirrus types

Stage 2: Tracking

Temporal Association

✓

Evaluated DeepSORT vs Norfair with 15s gaps

✓

MOTA: 43.3% (vs 11.3%)

✓

IDF1: 67.9% (vs 45.5%) - stable tracks

Stage 3: Attribution

Flight Matching

✓

Novel 4D→2D Methods: Geometric & Convolution

✓

71.75% overall attribution

✓

95-100% for short, 50-77% for thin

Flight Data Protected

Attribution visualization restricted due to EUROCONTROL security protocols

PythonDetectron2DeepSORTNorfairAzure DatabricksApache SparkComputer VisionPyTorch

Team Project with EquansAI Lead Developer

AI-Powered MEP Component Automation

Revolutionizing Modular Construction with Automated MEP Placement

>50%

Resource Savings

in MEP modeling time

Innovation in Construction AI

As AI Lead Developer, I designed and implemented a complete neural network system to automate Mechanical, Electrical, and Plumbing (MEP) component placement in modular construction projects. This proof-of-concept demonstrated how AI can drastically reduce manual MEP modeling time (currently consuming >50% of project resources) while improving accuracy and supporting scalable workflows.