Aref Azizian
AI Engineer at System Group, Ph.D. student at Amirkabir University of Technology
I build LLM systems — RAG pipelines, AI agents, and internal AI tooling that has made 400+ engineers more productive. My research is on learning from noisy labels.
- Now researching
- Learning from noisy labels in deep networks
- Now building
- RAG pipelines and AI agents at System Group
- Now reading
- Introduction to Linear Algebra by Gilbert Strang
Selected work
-
Journal article · Applied Soft Computing 2026
Contrastive siamese ensemble for high-fidelity label noise detection & correction: Theory & experiments
The journal version of my M.Sc. thesis: an ensemble of contrastive Siamese networks that detects and corrects instance-dependent label noise, with theory and experiments.
-
Paper · IEEE ICSPIS 2024
Preventing Overfitting on Noisy Labels Through Adaptive Checkpointing
An adaptive checkpointing method that keeps deep networks from overfitting to noisy training labels.
-
Open source · Python · MCP
Sub-Agent-MCP
MCP server for spawning sub-agents, enabling multi-agent workflows across AI tooling environments.
-
Article · Sep 2026 · In Persian
The Story of Hooshyar: Laying the AI Foundation with Open WebUI at System Group
How Hooshyar, System Group’s internal AI platform, grew from an Open WebUI deployment into the foundation for using AI across the software team.