A web-based platform for building machine learning pipelines with integrated tools for PFI, PDP, PCA, t-SNE, and autoencoders. Supports data visualization, explainable AI, and dimensionality reduction. Scales to large datasets via TU Dresden's HPC infrastructure.

Highlights

• 14+ ML models training directly in browser (no backend needed)
• Explainable AI suite: PDP, PFI, and feature contribution analysis
• Dimensionality reduction: PCA (biplots/scree), t-SNE, autoencoders
• Interactive visualizations: Parallel coordinates, correlation matrices, model insights
• HPC integration: Scales to large datasets via TU Dresden infrastructure
• Full client-side execution using Pyodide + TensorFlow.js
• Stack: Vue.js 3, Pyodide, Bulma CSS, Scikit-learn, Plotly.js, HighCharts, TensorFlow.js

A full-stack real estate management platform with a web frontend built in Vue.js, ASP.NET Core 8, and a SQL Server database.

Highlights

• ASP.NET 8 backend development for property listing platform
• RESTful API design with JWT authentication and authorization
• Azure cloud deployment: App Service, SQL Database, Key Vault
• Secure configuration management using Azure Key Vault
• Device-optimized image rendering service for faster page loads
• Entity Framework Core integration with advanced query optimization
• Role-based access control and secure API endpoint protection
• Automated CI/CD pipeline implementation for Azure deployment
• Stack: ASP.NET Core 8, ABP, Vue.js 2, SQL Server, Swagger UI, Bulma CSS, Bootstrap 4

Developed a CRUD application with TDD methodology.

Highlights

• Full CRUD operations implemented with Django
• one-time code login without traditional signup process
• Test-Driven Development methodology using pytest test suite
• Selenium automated browser testing for end-to-end validation
• Unit tests for models, views, and forms
• Stack: Django 5.0, Selenium automated test framework, pytest, SQLite

Explored the blackbox nature of machine learning algorithms by applying PDP,PFI, and SHAP model interpretability techniques to explain ML models' decissions.

Highlights

• Explored PDP, PFI, and SHAP for explaining black box models' predictions and enhancing transparency
• Simulated 8 different scenarios and visualized the methods' explanations with the true DGP
• Benchmarked the methods on 8 public datasets and 12 ML models
• Stack: scikit-learn, pandas, matplot, XGBoost

Used D3 to create Data Visualizations for public datasets as part of Data Visualization class at TU Dresden.

• Stack: D3.js

Improved model accuracy for cancer detection in whole slide images by designing a distance-aware Vision Transformer (ViT) that incorporated spatial relationships between patches into the self-attention mechanism.

• Stack: PyTorch

Using Deep Reinforcement Learning models to learn how to play Pac-Man based on interaction wiith the game and feedback received.

• Stack: PyTorch, Pygame