Automating the process of load finding for a company in the USA. Briefly, the main ideawas as follows: - Write scrapers for more than 30 websites that post loads. - Some of them were written with the Requests library - For the others, Selenium and Beautiful Soup were used - Create a Docker image for each scraper - Deploy the Docker images to AWS ECR - Create Lambda functions for each image and set up a scheduler for the scraping process to run the process once in a certain time intervalThe final output of each scraper was sent to the company's website backend.The entire project structure creation, development, AWS service

Researching, developing algorithms, collecting and labeling data, training models, solvingvarious computer vision tasks.Mainly worked with house interior and exterior images, as the company application wasrelated to real estate.A brief introduction to some of my projects:- The program receives as an input an indoor image of an apartment with a view fromthe window and the project was to detect on which floor the apartment is (+- 2 flooraccuracy was acceptable).- Match the query house images with other house images with the same postcode tofind the exact address of the query house. The difficulty of this project was that theimages of other apartments were distinct (photo angle, camera model, daytime, year,apartment renovation).- Finding duplicate listings in UK's 3 largest real estate websites.- Doing transfer learning for a model, or training from scratch if it is impossible toretrain. Familiar with different classification model architectures (VGG, ResNet,Inception) as well as object detection model architectures (YOLO, RCNN).

Wrote programs, developed algorithms and worked on projects worldwide for different industries and companies in following fields: image retrieval, image matching, image processing, object detection, image object detection and classification, semantic segmentation, etc.Some of the projects I've worked on:-Home image indexing for a city: Working on building image search engine for houses, where people could input an image and get similar images of the same house or building as an output using image feature detection, different descriptors such as SURF and KAZE (image retrieval). Also researching and testing many different image retrieval algorithms - DISK, Super Glue, PyTorch models, etc. - Human detection on sport related images: developing a program using sematic segmentation algorithms to be able to find sport players on different images. -Image grey scaling using image processing: Creating a system that converts and optimizes an Image conversion to greyscale for laser engraving using different image processing techniques and algorithms to increase the greyscale image quality, contrast, etc.Debugging and testing: Worked on debugging and testing of an algorithm of apartment images in order to increase image quality and remove all the defects such as dust on camera objective, aligning the image angles, straightening of images, sky replacement, etc.