Hi, I'm Tasos.
Machine Learning Engineer & Web Developer
Specializing in Python and Machine Learning. I also build fast, modern, and accessible web applications using React, Next.js, and Tailwind CSS.
Experience & Education
My academic and professional journey.
Law Office of Apostolos Tassis
AI Engineer & Digital Transformation Lead
- Specialized AI Legal Agent Development: Engineered a custom AI agent using Google Gemini models for the automated analysis and management of legal case files.
- RAG Infrastructure Implementation: Architected a high-performance Retrieval-Augmented Generation (RAG) system with a Qdrant Vector Store, indexing 50,000+ data chunks for instant legal research.
- Workflow Automation (n8n): Designed and deployed complex automated legal workflows, seamlessly integrating internal document archives with LLMs.
- AI Observability & Cost Tracking: Integrated Langfuse for real-time monitoring of performance, latency, and precise token-based cost analytics.
- Containerized On-Premise Deployment: Fully dockerized the AI stack to ensure local data hosting, prioritizing maximum security and GDPR compliance.
Starting with IT infrastructure oversight, I spearheaded the firm’s complete digital transformation, culminating in the design of a custom AI architecture. I developed a Retrieval-Augmented Generation (RAG) system that allows the firm to 'converse' with its own archive, reducing document retrieval time from hours to seconds. The entire infrastructure is Dockerized, ensuring that sensitive legal data processing remains local and secure. By implementing Langfuse, I achieved full transparency regarding AI cost-efficiency and model reasoning. My role now bridges the gap between traditional network security and the cutting edge of Generative AI.
Thesis: Machine Learning in Drug Discovery: Target Identification for COX-1 vs COX-2 Protein
University of Ioannina
View Code- Data Preparation: Cleaned 10k+ chemical samples, removed Invalid entries, filled missing values, and scaled descriptors for stable model training.
- Feature Engineering: Used 209 molecular descriptors and applied feature-importance analysis to identify which chemical properties influenced classification most.
- Model Development: Built neural networks (Keras Sequential & MLPClassifier) with ReLU hidden layers, sigmoid output, dropout, and L2 regularization to prevent overfitting.
- Training & Evaluation: Trained models with balanced classes and evaluated performance using accuracy, loss curves, confusion matrices, and prediction-truth comparisons.
- Interpretability: Applied permutation feature importance and heatmaps to quantify how much each descriptor contributed to COX-1 vs COX-2 classification.
- Pharmaceutical Impact: Demonstrated that optimized machine-learning models can reliably distinguish COX-1 and COX-2 inhibitors, supporting selective drug design and virtual screening.
In my thesis on target identification in drug discovery using machine learning, I focused on identifying viable molecular targets specifically for COX-1 and COX-2 enzymes. These enzymes are crucial in inflammatory processes, making them primary targets for developing anti-inflammatory drugs. Traditional methods for identifying these targets are costly and time-consuming, so my research aimed to develop a data-driven machine learning approach to streamline this process. Through SHAP analysis, I ensured model interpretability, aligning computational results with biological insights.
Internship: Machine Learning Engineer
EdenCore
- Data Annotation & Curation: Executed large-scale dataset annotation ensuring high-quality labels for computer vision tasks.
- Experimental Design: Designed and conducted controlled ML experiments, including hyperparameter tuning and ablation studies.
- Detection Pipeline Engineering: Implemented and troubleshot YOLOV5/YOLOV8 pipelines, optimizing training workflows.
- Foundation Model Utilization: Utilized SAM2, Florence-2, and Kosmos-2 for segmentation, captioning, and multi-modal analysis.
- Style Transfer Research: Conducted identity-preserving image manipulation research using StyleShot.
- Model Evaluation: Developed evaluation pipelines for object detection, computing mAP, IoU, and per-class performance.
During my internship at EdenCore, I worked on the full ML lifecycle, from large-scale data curation to model deployment. I optimized YOLOV5/V8 pipelines and experimented with SOTA foundation models like SAM2 and Florence-2. A key highlight was identifying and resolving a critical bug in an official GitHub repository for style transfer, which restored pipeline stability for identity-preserving experiments. This experience sharpened my ability to diagnose model failures and implement production-ready computer vision solutions.
ANASTASIOS TZARAS
Machine Learning Engineer & Web Developer
PROFESSIONAL EXPERIENCE
Law Office of Apostolos Tassis
March 2026 - Present
AI Engineer & Digital Transformation Lead
- Specialized AI Legal Agent Development: Engineered a custom AI agent using Google Gemini models for the automated analysis and management of legal case files.
- RAG Infrastructure Implementation: Architected a high-performance Retrieval-Augmented Generation (RAG) system with a Qdrant Vector Store, indexing 50,000+ data chunks for instant legal research.
- Workflow Automation (n8n): Designed and deployed complex automated legal workflows, seamlessly integrating internal document archives with LLMs.
- AI Observability & Cost Tracking: Integrated Langfuse for real-time monitoring of performance, latency, and precise token-based cost analytics.
- Containerized On-Premise Deployment: Fully dockerized the AI stack to ensure local data hosting, prioritizing maximum security and GDPR compliance.
Internship: Machine Learning Engineer
July - August 2024
EdenCore
- Data Annotation & Curation: Executed large-scale dataset annotation ensuring high-quality labels for computer vision tasks.
- Experimental Design: Designed and conducted controlled ML experiments, including hyperparameter tuning and ablation studies.
- Detection Pipeline Engineering: Implemented and troubleshot YOLOV5/YOLOV8 pipelines, optimizing training workflows.
- Foundation Model Utilization: Utilized SAM2, Florence-2, and Kosmos-2 for segmentation, captioning, and multi-modal analysis.
- Style Transfer Research: Conducted identity-preserving image manipulation research using StyleShot.
- Model Evaluation: Developed evaluation pipelines for object detection, computing mAP, IoU, and per-class performance.
EDUCATION
Computer Science and Engineering
2019 - 2024
University of Ioannina, Greece
Thesis: Machine Learning in Drug Discovery: Target Identification for COX-1 vs COX-2 Protein
- Data Preparation: Cleaned 10k+ chemical samples, removed Invalid entries, filled missing values, and scaled descriptors for stable model training.
- Feature Engineering: Used 209 molecular descriptors and applied feature-importance analysis to identify which chemical properties influenced classification most.
- Model Development: Built neural networks (Keras Sequential & MLPClassifier) with ReLU hidden layers, sigmoid output, dropout, and L2 regularization to prevent overfitting.
- Training & Evaluation: Trained models with balanced classes and evaluated performance using accuracy, loss curves, confusion matrices, and prediction-truth comparisons.
- Interpretability: Applied permutation feature importance and heatmaps to quantify how much each descriptor contributed to COX-1 vs COX-2 classification.
- Pharmaceutical Impact: Demonstrated that optimized machine-learning models can reliably distinguish COX-1 and COX-2 inhibitors, supporting selective drug design and virtual screening.
LANGUAGES
- Python
- TypeScript
- JavaScript
- SQL
- C++
- Java
LIBRARIES & AI
- LangChain
- Google AI
- React
- Next.js
- FastAPI
- PyTorch
- TensorFlow
- Chainlit
TOOLS
- Docker
- n8n
- Langfuse
- Qdrant
- PostgreSQL
- Git
- Linux
CERTIFICATIONS
- Scientific Computing with Python
- Deep Learning with Python
- ML with Python
REFERENCE
Paris Amerikanos, Machine Learning Engineer - AI Lead, Eden Library.
"I had the pleasure of mentoring Tasos during his 2-month internship at Eden Library, where he worked as a Machine Learning Engineer. He quickly adapted to our projects and exceeded our expectations within the first week. Tasos successfully ran and completed three computer vision experiments, showing strong technical skills and great attention to detail. He’s easy to work with, communicates well, and was always eager to learn. His potential in ML & R&D, is clear, and I’m confident he’ll excel in future roles. I highly recommend him for any opportunity in the field!."
Built with Next.js & Tailwind CSS | Dynamic Portfolio by Tasos Tzaras.
Featured Projects
A selection of recent work in Front-end and Full-stack development.
Weather App
Real-time weather forecast application using OpenWeatherMap API. Supports global city search with live data.
Currency Converter
Currency converter tool with live exchange rates. Automatically calculates conversions based on data from an external API.
Movies Database Extraction
Movie data extraction and processing script. Handles large datasets and organizes them into a structured JSON format.
City Explorer
A real-time landmark and restaurant recommender system based on user location. Utilizes Foursquare API for data fetching and Geolocation API for spatial awareness.