Education · Technical redirection · Software engineering
From elite engineering foundations to AI-focused software systems.
I ranked 367 in Konkur, studied Aerospace Engineering at Sharif University of Technology, and later left that direction because I did not want my engineering path to be connected to missile or weapon-system work. I redirected my education toward AI and software, and I am now studying Software Engineering at Azad University with a 19.10 GPA.
Konkur rank: 367Sharif University of Technology · Aerospace backgroundSoftware Engineering · Azad UniversityGPA: 19.10
World-class curriculum · Books · Self-study
Learning AI through serious courses and foundational books.
My learning path is built around strong global resources: Machine Learning Specialization and Deep Learning Specialization by Andrew Ng, the DeepLearning.AI TensorFlow Developer Professional Certificate by Laurence Moroney, Harvard CS50x Introduction to Computer Science, and MIT 6.S191 Introduction to Deep Learning.
Machine Learning with PyTorch and Scikit-Learn · Sebastian Raschka
Deep Learning with PyTorch · Eli Stevens, Luca Antiga, Thomas Viehmann
Introducing MLOps · Mark Treveil and contributors
Build a Large Language Model (From Scratch) · Sebastian Raschka
Natural Language Processing with Transformers · Lewis Tunstall, Leandro von Werra, Thomas Wolf
Regression · Boosting · CV · NLP · Transformers
A research path from tabular models to transformer systems.
My project work started with simple regression and classical ML, then moved into gradient boosting models, medical diagnostics, feature engineering, and computer vision. I have worked with fine-tuning and transfer learning using models such as YOLOv8, EfficientNet, and Inception-style networks. In NLP, I moved from Disaster Tweets and SMS classification into Quora Question Pairs, LSTM with attention, self-built cross-attention, transformer encoder design, BERT fine-tuning, XLM-style multilingual work, GPT-2 experiments, and the broader Hugging Face transformers ecosystem.
Regression & boostingYOLOv8 · EfficientNet · InceptionQuora Question Pairs · attentionBERT · XLM · GPT-2
APIs · Typed frontends · Databases · Automated delivery
Turning ML models into real applications.
I have experience moving from notebooks into working systems: ONNX Runtime inference behind FastAPI, React and TypeScript interfaces built with Vite, PostgreSQL persistence through SQLAlchemy and Alembic, Docker and Nginx services, and automated delivery to Render and Neon through GitHub Actions.
React · TypeScript · ViteFastAPI · ONNX Runtime · DockerPostgreSQL · SQLAlchemy · AlembicNginx · Render · Neon · GitHub Actions
Automation · Experiment tracking · Event-driven thinking
Adding engineering discipline around the model lifecycle.
I have used MLflow for experiment tracking in projects, Airflow for automation, and Kafka in the lung disease detection app to simulate local event-driven tasks. My MLOps direction is focused on reproducible training, reliable inference, automated workflows, model logging, and production-style system thinking.
MLflow trackingAirflow automationKafka simulationReproducible ML workflows
Programming · Frontend · Systems basics
Beyond Python: lower-level logic and typed web interfaces.
Python is my main AI language, but I also have experience with C, C++, Swift, Kotlin, JavaScript, TypeScript, HTML, and CSS. I designed a local database-like application in C++, and shipped the same offline vocabulary app twice from one Windows machine — once in Swift and SwiftUI, then ported screen for screen to Kotlin and Jetpack Compose — with no Mac, no Android Studio and no local simulator or emulator, every build produced by cloud CI. I now use React and Vite to build typed, responsive, accessible interfaces that connect directly to deployed APIs.
C programmingC++ local database-style appSwift · SwiftUI offline iOS appKotlin · Jetpack Compose Android portJavaScript · TypeScript · ReactResponsive and accessible frontend systems
Current stage · Full-stack ML apps · Transformers ecosystem
From end-to-end medical AI to full-stack ML application engineering.
After building the lung disease detection pipeline, I extended that experience with an SMS spam classifier built as a complete split-service application: ONNX Runtime inference behind FastAPI, Fernet-encrypted PostgreSQL persistence through SQLAlchemy and Alembic, and a React and TypeScript interface served by Nginx. GitHub Actions runs backend, frontend, browser, accessibility, and deployment-smoke checks before migrating Neon and triggering Render deployments. I am continuing to study transformer model families, the Hugging Face ecosystem, and LLM foundations.
Lung disease detection appSMS spam classifierReact · TypeScript · FastAPIPostgreSQL · Alembic · CI/CD
Future direction · LLMs · RAG · Agents
The next layer is advanced AI systems engineering.
My future work is focused on mastering the transformers ecosystem, building an LLM from scratch, deploying more complete AI apps, diving into RAG and reasoning techniques, and mastering agentic workflows. The goal is to keep moving from isolated experiments toward advanced, reliable, and explainable AI systems.
Transformers ecosystemLLM from scratchRAG & reasoningAI agents
