Projects
A curated portfolio of end-to-end builds and research experiments—focused on practical AI systems, data pipelines, and full-stack projects that prioritize reliability, evaluation, and real-world constraints.
Dataset engineering for a Macedonian LLM. Lots of prompt crafting, format fixing, and convincing a model to speak like a real human.
Local NotebookLM-style assistant with strict citations and abstention. If the docs don’t support it, it won’t answer. Simple rule, surprisingly hard for the LLM.
CLIP-style multimodal dementia classifier using speech and transcripts. Training loops, evaluation pipelines, and lots of GPU fan noise.
Data pipeline + LSTM prediction stack for 1k+ coins. Ingest, model, visualize, repeat. Markets still irrational, but now at least graphed.
Interactive algorithm visualizer so sorting algorithms can’t hide their secrets. Made for learning, debugging, and satisfying animations.
Unity/C# fighter where the real final boss was the physics engine.