Mahdi Bideh

Computer Vision & AI Engineer

I turn hard visual problems into deployed systems — from foggy-image dehazing and object detection to pattern recognition and LLM-powered AI workflows.

PythonPyTorchOpenCV C++Deep LearningPattern Recognition DockervLLMGolangFastAPI Nest.js
View projects mahdi.76911@gmail.com

About

I'm a Computer Vision and AI Engineer with an M.Sc. in Computer Science (Artificial Intelligence) from Amirkabir University of Technology and 5+ years across AI research, production backend, and mobile development. I like working where research meets shipping — from a Master's thesis in foggy-image dehazing with 1 conference paper, to real-time stock platforms, agentic AI workflows, and LLM serving at scale.

5+years in software & AI
1conference paper (plate recognition)
10+dehazing methods benchmarked
8+products led to cloud deployment
Dehazing methods compared side by side on a foggy KITTI street scene
My method vs. classical & deep baselines on a foggy KITTI street scene — haze removed while lane and vehicle structure stays intact.
Synthetic fog generation pipeline with depth map, fog-density maps and composited foggy scenes
Synthetic fog generation pipeline: a depth map is shaped by β into a fog-density field and composited with intensity A — thousands of clean/foggy training pairs from one scene.

Projects

Selected work — research and production, 2020 → present. Full details in my portfolio.

Research · Deep Learning

GBDA Net — Dehazing with Object-Visibility-Aware Loss

Master's thesis: single-image dehazing autoencoder trained with IoU of YOLOv5 detections, improving vehicle detection in fog by 22% for ADAS.

PyTorch · YOLOv5 · KITTI · RESIDE
Research · CV
Taxonomy of image defogging methods: image restoration, image enhancement and fusion-based approaches

Classical Dehazing Methods Survey

Reproduced DCP, Retinex, CLAHE, Fattal and Tarel from the papers with per-method contrast metrics (SSIM, PSNR, UQI, no-reference) in MATLAB + Python.

MATLAB · Python · OpenCV
Deep Learning · CV
YOLO car detections on a KITTI scene, with and without dehazing

YOLOv5 on KITTI

Custom data configs, KITTI→YOLO label conversion, trained weights — detection backbone for the dehazing evaluation pipeline.

YOLOv5 · KITTI
Deep Learning
Persian license plates: reference plates and synthetic test variations

Persian License Plate Recognition

89%+ accuracy under challenging conditions with YOLO, CAPSNET and U-Net on a synthetic dataset of 10,000+ plates. Paper submitted.

YOLO · CAPSNET · U-Net
Research · Synthetic Data
Aerial CARLA simulator street scenes at increasing fog density

Forging Fog — CARLA Synthetic Fog Dataset

Thousands of clean/foggy street-scene pairs generated in the CARLA simulator with controlled depth and fog density, extending the dehazing training pipeline beyond real-world KITTI/RESIDE.

CARLA · Python · OpenCV
Backend · Golang

Sahmnak — Real-Time Stock Simulator

High-performance Golang crawler syncing the Iranian stock exchange in seconds; MQTT + WebSocket streaming to thousands of concurrent users.

Golang · MQTT · WebSocket
Backend · Node

Bookitall Microservices

Microservice architecture for an online bookstore on Nest.js: REST APIs, PostgreSQL/MongoDB, automated Docker deployment.

Nest.js · Docker · PostgreSQL
Mobile

Sahmnak Android App

Kotlin app with real-time price updates over WebSocket and a lightweight in-app cache; Java→Kotlin legacy refactor.

Kotlin · WebSocket
Research · NLP/ML

Board Game Recommender Systems

Two recommender systems (similar-game and questionnaire-based) on data from ~500 participants — dataset, questionnaire, and models.

Python · Scikit-learn

Skills

The tools I reach for, by category.

ProgrammingPython, C++, C# (+WPF), MATLAB, R, Java, Kotlin, Golang, SQL, JavaScript, PHP
Deep Learning & CVPyTorch, TensorFlow, Keras, Scikit-learn, OpenCV, YOLOv5, image quality assessment (SSIM, PSNR, UQI, no-reference descriptors)
AI EngineeringAgentic AI workflows, LLM serving with vLLM, pattern matching pipelines, model deployment
Backend & APIsFastAPI, Flask, Nest.js, Django, REST, WebSocket, MQTT, PostgreSQL, MongoDB
DevOps & ToolingDocker, Docker Compose, CI/CD, Git, Linux (Ubuntu), AWS, Jupyter, LaTeX, CARLA Simulator
LanguagesPersian (native), English (advanced — TOEFL 106)

Experience

Research and industry, in parallel.

Get in touch

Open to computer vision, AI engineering, and research opportunities.

mahdi.76911@gmail.com github.com/mahdi76911 LinkedIn