Python Developer

Kandidát je inženýr počítačového vidění, který se zajímá především o umělou inteligenci. Co se týče jeho komerčních zkušeností, pracuje s jazykem Python téměř 2 roky. S jazykem Python však pracoval již během celého studia. Ve svém současném zaměstnání je zodpovědný za návrh modelů, prediktivních modelů, které jsou založeny na vizuálních datech. Dále zpracovává snímky pocházející z kamery na základě funkce, kterou zákazník potřebuje. Například píše pro nevidomé a slabozraké, aby mohli číst. Kromě toho dělá další věci v oblasti umělé inteligence, jako je detekce objektů, detekce textu, rozpoznávání, optimalizace a tak dále. Kromě Pythonu používá knihovny jako Scikit, Tensorflow, PyTorch, Keras. Co se týče Djanga, občas ho používá pro demonstraci, ale má s ním roční komerční zkušenosti. Co se týče AWS, jeho zkušenosti jsou nyní dvouleté. Kromě toho vždy pracoval s agilními metodikami, jako je Scrum. Baví ho také vést tým, proto je otevřený vedoucím povinnostem v budoucnu.

• Dostupnost v polovině července/srpna

Computer Vision Engineer (Dates: Unknown)

• Led a project to develop a real-time document guidance feature for the blind and visually impaired using image segmentation.
• Managed data collection and annotation pipelines on a large scale.
• Customized the model architecture to increase on-device inference speed by 1000%.
• Implemented a mask consistency algorithm and customized loss function to reduce False Positives by 14%.
• Re-implemented a text detection model from PyTorch to Tensorflow 2.x for on-device detection of incidental text, resulting in a 400% increase in inference speed.
• Applied weight pruning to reduce model size by 20%.
• Implemented face detection and alignment using BlazeFace for Mobile GPU inference on Google Glasses.
• Implemented face recognition module based on MobileNetv3 feature extraction and Cosine Similarity.
• Implemented a text orientation classifier for a better reading experience for the visually impaired and blind.
• Implemented real-time 1D barcode detection using traditional computer vision methods.

Principle AI Engineer (Dates: Unknown)

• Managed a team of 3 to implement data in a graph database platform (neo4j).
• Designed data models for a large amount of unstructured data and exported them to a graph database.
• Managed a team to design a tool for synthetic data generation.

Computer Vision Engineer (Dates: Unknown)

• Implemented a text localization Neural Network to improve the accuracy by 5% (mAP) compared to the previous solution.
• Implemented and benchmarked a language agnostic text recognition network.
• Designed and implemented an API and web platform for a complete text detection platform.
• Implemented an API with Django for measuring image similarity based on MobilenetV2 features and cosine similarity metric.
• Benchmarked a large-scale image dataset on GCP cloud vision and AWS Rekognition to improve production accuracy.
• Improved the speed of API calls by 50% for the benchmark by using Asynchronous Python with AsyncI/O.

Backend Developer Intern (Dates: Unknown)

• Developed a web API based on Django-Rest framework.
• Debugged data synchronization between the web platform and the Android app.

Publication Record:

• Accepted paper "AI-assisted image analysis of cervical spine motion X-Ray images" at EuroSpine 2022.
• Published paper "t-EVA: Time-Efficient t-SNE Video Annotation" at ICPR 2020.

Education:

• Master of Embedded Software Engineering, Delft University of Technology (Dates: Unknown)
  • Specialized in Software and Networking with a Computer Science orientation.
  • Master Thesis in Computer Vision and Deep Learning.
• Bachelor of Electrical Engineering, Shahid Beheshti University (Dates: Unknown)
  • Specialized in Electronics.
  • Ranked in the top 10% among the students in a class of 40 students.

Projects:

• Master Thesis at TU Delft (Dates: Unknown)
  • Conducted video classification on activity recognition datasets using various models.
  • Implemented a time-efficient video annotation tool using T-SNE and features similarity.
  • Increased annotation speed by 10 times compared to trivial video annotation tools.
• Advance Practical IoT (Dates: Unknown)
  • Implemented a real-time License Plate Recognition Android app and web API in a team of 2 people.
  • Used YOLOv3 object detection and network compression techniques in TensorFlow.
• Deep Learning (Dates: Unknown)
  • Implemented YOLOv2 for object detection and localization in a team of 5 people using PyTorch.
  • Applied network compression techniques for improving the speed of the detection process.
• Intelligent User Experience Engineering (Dates: Unknown)
  • Designed and implemented a support system for patients with diabetes type-2 on a Pepper humanoid robot using NAO SDK.
  • Demo project at Fruitpunch AI event with TNO collaboration about Hybrid Intelligence and Explainable AI.

Languages:

• English: Fluent (Toefl Score: 100)
• Farsi: Native