Jan Kociniak Email & Phone Number

Building next-gen AI assistants

Amsterdam, North Holland, Netherlands

• Developed a novel deep learning algorithm for modeling Riemannian geometry of high-dimensional point clouds as a part of the Master thesis.• Gained practical experience with 3D point clouds and meshes processing and visualisation.• Gained research experience with implicit neural representations, manifold learning and shape reconstruction.• Presented at 2024 Maths4DL Geometric Deep Learning workshops at the University of Cambridge.• Supervised by Willem Diepeveen (Cambridge Image Analysis Group) and Erik Bekkers (Amsterdam Machine Learning Lab).

Poland

• Implemented the prototype of an anomaly detection pipeline for CCTV data based on normalizing flows.• Implemented the pose estimation module working in real-time for highly crowded scenes.• Implemented the gait cycle regression module using spatio-temporal graph neural networks.

• Designed and implemented a pipeline to construct device trajectories from raw geofencing signals.• Built interactive visualisations for extracted trajectories using Jupyter Notebook, HoloViews and folium.

• Cleaned datasets and wrote automation scripts for the eye-tracking AI model.

• Graded problems at all levels.

Master Of Science - Ms, Artificial Intelligence

Focused on unsupervised geometric deep learning. Deep Learning 2 project: Random Walks on CW Complexes for Graph Positional Encodings.

Bachelor'S Degree, Computer Science

Thesis topic: Hierarchical Risk Parity algorithm for portfolio optimization.

Bachelor'S Degree, Mathematics

Finished all courses with GPA 4.02/5, did not submit thesis due to burnout. Thesis topic: Comparison of scaling laws of hyperbolic.