Amirali Hosseini Abrishami
Cybersecurity Researcher & Systems Developer
Education
Thesis: GraphRAG-based architectures for knowledge-intensive ML pipelines.
Experience
Front-end development internship. Built and maintained Angular components for client-facing web applications within an agile team.
Skills
- Languages
- C · Python · Bash · GDScript · (Rust — in progress)
- ML / AI
- PyTorch · PyTorch Geometric · HuggingFace Transformers · cuML · scikit-learn · Graph Neural Networks
- Systems
- Linux · POSIX APIs · Docker · GNU Make · systemd
- Game Dev
- Godot Engine 4 · Unreal Engine 5.1 (Blueprints)
- Tools
- Git · GitHub Actions · VS Code · Wireshark · strace
Selected Projects
Graph neural network (GCN, SkipGCN) vs. tabular baseline (Random Forest) comparison for illicit transaction classification on a real-world AML dataset. Accelerated tabular training with GPU via cuML. Random Forest achieved F1: 0.80 (Precision: 0.90, Recall: 0.72); GNN track revealed limits of graph structure without rich neighbourhood features.
3-class semantic segmentation pipeline (background, border ring, plate interior) using a U-Net architecture. Achieved 0.9982 pixel accuracy on the test set.
Browser-based solo tabletop RPG with an LLM as Game Master. FastAPI backend with provider-agnostic LLM routing (OpenAI, Anthropic, Gemini, Ollama). Persistent memory via rolling summarisation every 10 turns; React 19 + Vite frontend.