CV

Master

• Artificial Intelligence.
• Seminar
  • Generative Adversarial Networks (GAN) Survey

Bachelor

• Electrical and Electronics Engineering

Team Lead & Python Developer

• Prototyped 4 services that were responsible for photogrammetric processing on drone images.
• Architectured a photo aesthetic assessment service which was capable of visual RAG to analyse over 60 parameters with gathered specialized dataset.
• Introduced and programmed a graph-based workflow to create a user-defined dialogue flow for a personalized chatbot that can be customized with LLMs and RAGs.
• Leveraged asynchronous I/O to enhance server capacity and reduce response times.
• Improved multi-processing, multi-threading, and asynchronous programming skills of 12 of my coworkers.

Machine Learning Engineer

• Optimized RAM and CPU of the camera streaming service by about 30%.
• Implemented a streaming service that can connect to 25 HD cameras with 12 cores CPU and 2.5 GB RAM to proceed 20 frames per second.
• Integrated a system for detecting personal protective equipment with 4 cameras capable of 5 categories.
• Developed a task-oriented framework to deal with ML models and pipelines in the production environments.
• Dockerized 3 AI and 5 Backend services to run reliably and securely in any infrastructure.
• Trained 5 of my co-workers to lead the different product services.

Machine Learning Engineer

• Designed a pipeline containing 5 ML models (object detection, layout composition, pose detection, depth estimation, and image blending) for human localization on the background image.
• Trained an ACGAN model on the heart disease dataset to augment the data which achieved 70% accuracy in generating fake data.

Internship

PhotoGear

• Orthomosaic photo generation from drone images with a pipeline of radiometric calibration, Structure from Motion (SfM), and Orthomosaic Generation.
• Features
  • NDVI calculation and vegetation analysis
  • Structure from Motion (SfM) for 3D reconstruction with COLMAP and OpenSFM
  • Orthomosaic Generation with GDAL and PDAL
  • API Gateway and Queue management with FastAPI for data processing and task management

Photo Aesthetic Assessment

• A platform to assess photo aesthetic parameters with LLM combined with VRAG and some computer vision pipelines.
• Features
  • Visual-RAG to retrieve photo and the context with Langchain
  • MongoDB to search on embedding vectors
  • Async FastAPI endpoints
  • Clean architecture Design
  • Dependency injection and wiring with Lagom

Parstech Video Intelligent Assistant

• AI assistance to process and analyze videos with modules like human detection, face recognition, and license plate recognition. I deeply worked on human detection and restriction area application, Camera handling and streaming, DevOps, and microservices of this application.
• Features
  • Human detection and tracking
  • Video Processing with OpenCV
  • FastAPI implementation for APIs
  • Gstreamer and OpenCV cores for the Camera management service
  • Task queueing and workflow management with Celery
  • Containerized services
  • In-Memory data transfer with Redis and python shared memory
  • Multi-Processing and Multi-Threading features implemented

LipReading model on Persian Dataset

• Train a LipReading model on the Persian dataset.

Personal Protective Equipment (PPE)

• A program to identify workers’ safety equipment in workshop and construction environments, such as gloves, helmets, glasses, masks, safety vests, warning capabilities for people who enter prohibited areas, fire detection, and identification of work tools.
• Features
  • Human detection and tracking
  • Pose estimation to check human body status
  • Body part localizer for head, hands, and chest
  • Equipment classifiers models
  • Pipeline implementation to estimate human fall
  • Faster functions with numba jit compiler

Pressure Ulcer

• Analysis of pressure sensor data of hospital beds for prediction and diagnosis of bed sores.
• Features
  • Body Segmentation Models.
  • Pose estimation model for pressure sensor data
  • Video processing with OpenCV
  • Signal Capturing from human poses

Cardiac Medical Data Augmentation

• Train ACGAN on real Cardiac Medical Data in order to generate fake data for augmentation.

Human localization

• In two categories of foreground and background images, it takes a personal image from the foreground, and in order to place it in the background image, it first finds the right place then masks the image and places it in the background, and then corrects the color and light also improves the image by a model.
• Features
  • Depth Estimation
  • Image brightness correction & Color correction models
  • Object detection models to find humans and other objects in the Image.
  • Image Perspective calculation
  • Image blending

OCR

• Optical Character Recognition on German administrative forms.
• Features
  • Use tesseract and self-trained text recognition neural network.
  • Use Clustering algorithms to find lines.
  • Use OpenCV and Image processing methods to find information blocks.