- Implemented a prediction system for OHCL stock values based on a Transformer architecture.- Conceptualized and Integrated a reinforcement learning agent for stock trading. For the training, I build a custom GYM environment and used Proximal Policy Optimization(PPO) to fit an actor-critic model.

- Integrated and build a component based on a custom yolov5 architecture, for real-time object detection.

- Implemented a real-time semantic segmentation system based on DDrnet architecture for fruits anomaly.- Performed extensive analysis of the quality of the training data.- Deployed the system on AWS EC2 machines.

- Developed an ETL pipeline for valuable parcels in Berlin.- Implemented a system for generating large datasets (>100 GB) for land cover segmentation for different areas of the world and automatically store it on S3 Buckets.- Created a pipeline for training multiple models based on vision transformers, for the task of land use segmentation from RGB satellite images.- Integrated GitLab CI/CD pipelines to run automatically run custom training.- Developed an extraction pipeline for buildings and roads from open street maps. It takes <10 mins per USA state.

- Implemented Live Streaming desktop application using QT and OpenCV.- Implemented real-time vehicles detector using YOLOv3.- Researched and tested several image enhancement methods ( Super Resolution,Denoising, Deblurring) Such as FFDNET, DnCNN, DIDNN, EDVR and deblurGAN.Optimized the execution time of the tested models.

Created a dataset of 100k faces with age as a label by merging several datasets.Finetuned Resnet50 for age estimation and densenet121 for gender detection.keywords: Python - FAISS - OpenCV - Tensorflow - ZMQ sever