I'm a Computer Engineering student building toward a career founding my own AI company. I currently serve as Co-Founder & Technical Lead at AI Builders Iran, where I set technical direction across our projects and lead the LLM/RAG engineering track while our team builds computer vision systems for real industrial and medical problems.

My work sits at the intersection of two disciplines: medical & industrial computer vision (segmentation, detection, deployment) and LLM/RAG engineering (retrieval pipelines, local model serving, agentic systems). I don't just prototype — I take projects from architecture design through training, bug-fixing, and containerized deployment, and I document the process for the community through a Persian-language educational series on LLM fundamentals.

Currently:

• 🧠 Preparing my brain MRI tumor segmentation project for academic submission (JOSS / arXiv, cs.CV) — adding baseline comparisons and an ablation study
• 📚 Progressing through a structured RAG curriculum — Document Processing → Retrieval → Advanced Retrieval → Generation → Production Systems
• 🏗️ Directing LLM integration on AI Builders Iran's Safety Monitoring System as Technical Lead
• ✍️ Publishing a 25+ episode Persian LLM-fundamentals series on LinkedIn (tokenization → attention → transformers → fine-tuning)
• 🌐 Building out my personal RTL Persian portfolio (React, TypeScript, MDX, KaTeX)

Co-Founder & Technical Lead — AI Builders Iran Co-founded and technically lead a team building applied AI systems from the ground up. Responsible for architectural decisions across projects, dividing work along clear technical boundaries (e.g., detector/corrector, data pipeline/modeling), reviewing implementations, and driving projects from prototype to deployed, containerized products. Lead the team's LLM/RAG engineering track personally.

🧠 Brain MRI Tumor Segmentation — BRISC 2025 ResNet34 U-Net3+ architecture with Boundary Loss for precise tumor boundary segmentation. Built a leakage-safe data pipeline (threshold selection moved from test to validation set), aspect-ratio-preserving letterbox preprocessing, full checkpoint resumption across 40+ training epochs, and a composable OpenCV/matplotlib visualization & reporting suite (bounding boxes, overlays, convex hulls, centroids). Deployed as a Streamlit + FastAPI + Docker stack. Currently being prepared for peer-reviewed publication (JOSS, arXiv cs.CV) with baseline comparisons and ablation studies.

🦺 Safety Monitoring System — AI Builders Iran A full CV-to-LLM pipeline for industrial hazard reporting: YOLOv8 for real-time hazard detection, feeding into a locally-served Qwen2.5-1.5B-Instruct model that generates structured HSE reports through a Jinja2-templated prompt engine and rule-based severity engine. Shipped end-to-end with a Gradio interface. Led the LLM integration layer as Technical Lead.

🔧 PCB Defect Detection Two-stage YOLO pipeline for industrial PCB inspection, with GPU-aware inference, PyTorch-native model serving, and a clean Streamlit interface built for inline QA review by non-technical operators.

📊 ML_Lablator A multi-page Streamlit application covering the full ML practitioner workflow — data cleaning, EDA, feature engineering, model training, and evaluation — with clustering visualizations, confidence-band learning curves, and one-click exportable charts. Fully documented, packaged, and open-sourced.

🩺 medvision (in progress) Open-source library for MRI/CT artifact detection and correction. A single pipeline interface (MedVisionPipeline.run()) handles NIfTI, DICOM, and standard image formats for both single-image and batch-folder processing — designed with a clean detector/corrector boundary for collaborative development.

Languages & Core

ML / DL / CV

LLM / RAG

Deployment & Tools

Learn deeply. Build consistently. Ship things that work.