Projects | Andreas Avgoustis

Machine & Deep Learning

Models and Applications for Lip Reading and Voice Reproduction

The Challenge:

Building a multi-modal temporal model to interpret lip movements from video frames without audio.

Architecture:

Utilized Visual Language Model (VLM) stacked with Retrieval-Augmented Generation (RAG) and Greek Large Language Model for sequence modeling.

Tech Stack:

Python PyTorch OpenCV CUDA

Code

Single Cell RNA Seq Classification

The Challenge:

Classifying brain cell lineages (such as Astrocytes, Neurons, Microglia, etc.) from high-dimensional single-cell RNA-sequencing (scRNA-seq) data.

Architecture:

Developed a Joint Autoencoder & Classifier that operates on ~16,678-dimensional gene expression profiles.

Tech Stack:

Python TensorFlow NumPy Scikit-Learn

Code

Optimization of Healthcare Process Management using Machine Learning

The Challenge:

Managing asynchronous streaming structures inside healthcare systems without encountering data starvation bottlenecks.

Architecture:

Constructed a high-throughput, non-blocking asynchronous processor leveraging virtual threads and structured data access wrappers.

Tech Stack:

Python TensorFlow XGBoost

Code

Identifying Earthquakes in Low-Cost Sensor Signals Contaminated with Vehicular Noise

The Challenge:

Filtering out heavy structural and environmental traffic noise arrays from low-cost seismic sensors to accurately classify seismic anomalies.

Architecture:

Implemented short-time customized Deep Convolutional Networks for spectral signal filtering.

Tech Stack:

Python TensorFlow ObsPy

Code

Dermoscopic Melanoma Image Classification

The Challenge:

Training a robust model capable of diagnosing dermoscopic images with unbalanced class distributions.

Architecture:

Built an ensemble architecture using pre-trained EfficientNet and Vision Transformer (ViT) nodes with focal loss optimization layers.

Tech Stack:

Python TensorFlow Computer Vision

Code

System Programming

Crypto Analytics & Prediction Platform

The Challenge:

Orchestrating real-time data streaming architectures with strict low-latency ingestion barriers.

Architecture:

Designed a distributed streaming node engine linked to concurrent analytical workers executing rapid time-series predictive modeling.

Tech Stack:

Java Spring Boot Python FastAPI PostgreSQL React Tailwind CSS

Code

Real time Monitoring System & Feature Engineering

The Challenge:

Executing intensive on-the-fly mathematical operations on telemetry data streams without impacting pipeline throughput.

Architecture:

Architected a modular parallel data ingestion engine featuring dynamic windowing configurations for custom feature calculation.

Tech Stack:

Python FastAPI NGINX Grafana SQLite

Code

Intelligent Tax Filing

The Challenge:

Parsing intricate legal constraint trees into deterministic software rule sets safely.

Architecture:

Engineered a highly secure rule engine managing state transitions with zero external side effects and verifiable validation audits.

Tech Stack:

Python OpenAI FastAPI Docker React Shadcn UI Tailwind CSS

Code

Tutoring System

The Challenge:

Architecting an intelligent, personalized data relationship structure mapping real-time learning curves.

Architecture:

Formulated a clean object-oriented data access engine deploying strict relational entity mappings for instant performance scoring.

Tech Stack:

Java 25 Spring Boot 4 Hibernate PostgreSQL React Shadcn UI Docker

Code