I'm Sebastian Frey

Developed a novel spatiotemporal framework using LLM-derived semantic graphs to predict Cost-Per-Click (CPC) auctions.Proved that explicit relational structures improve medium-to-long-term robustness over foundation models.

Architected a 'zero-caching' real-time pipeline for personalized news-to-podcast generation. Orchestrated complex LLM workflows and Google Cloud TTS synthesis for high-fidelity unique audio output.

Comparative manifold learning pipeline applying K-Means, GMM, and HDBSCAN to a 28-dimensional financial dataset projected into 3D latent space via UMAP to isolate fraudulent transaction topologies.

Leveraged LLMs to create source-aware unbiased news articles from idiologically oposed sources (Fox vs. CNN). Works independet of language and topic to create a news dashboard of unbiased information

A real-time multiplayer chess game powered by WebSockets. Server validates moves with python-chess and broadcasts to all players instantly. Features a round-robin queue system, move animations, and persistent game state.

An end-to-end ML pipeline for high-throughput churn prediction. Features automated preprocessing, GridSearch hyperparameter optimization for XGBoost, and production-grade validation strategies to forecast customer lifetime value (LTV).