Tom Kre Sal
Verified Expert in Engineering
Machine Learning Engineer and Developer
Tom is a full-stack machine learning engineer with several years of experience in machine learning and software engineering, working on projects driven by smart algorithms and data. Tom gets a kick out of creating solutions to complex problems that positively impact the lives of others. Tom likes to keep abreast of cutting-edge technologies, which he then applies to production-ready systems—giving his clients an advantage over their competitors.
Artificial Intelligence (AI), Computer Vision, Machine Learning, Back-end, Python, Linux, Cloud, Generative Pre-trained Transformers (GPT), Natural Language Processing (NLP), GPT, Data Science
The most amazing...
... thing is to imagine something that does not exist and then try to build it.
Data Scientist/Data Engineer | Receipts Data Extraction Application
Sanad Cash, Inc
- Designed and implemented a machine learning system that recognizes and extracts meaningful information from accounting documents using deep learning algorithms.
- Designed and implemented highly scalable, reliable, and distributed back-end services using Python, FastAPI, Celery, Redis, and MongoDB to process incoming data in real-time.
- Implemented the detection and classification of important objects on input documents, like QR codes, tables, headings, company logos, signatures, and many more.
Computer Vision Developer
- Architected and built a pipeline to intelligently extract 3D-scene data from sequences of panoramic images, using a mix of geometric and deep learning approaches.
- Created custom neural network architectures for object detection, semantic, and instance segmentation designed to detect different construction-specific objects and their changes over time.
- Implemented the detection and anonymization of human faces and bodies on input video feeds for privacy reasons.
- Reconstructed the visual SLAM algorithm's walking trajectory from the input video feed to properly align the projected path on the floor plan, including important features (like walls, rooms, doors, windows, etc.) on building drawings.
- Managed the complete lifecycle of ML experiments (labels management, experiments tracking, evaluation, and deployments) with third-party tools like Google Vertex AI, Weights & Biases, Amazon SageMaker, or Metaflow (an open source project).
Lead Full-stack Machine Learning Engineer
- Developed a solution to extract meaningful information from invoices based on computer vision, natural language processing, and machine learning in general.
- Trained a conditional random field model on named entity recognition to classify important invoice fields.
- Created a solution to extract text from invoice images with a custom optical character recognition system based on deep learning.
- Designed the solution to be a highly scalable, robust system with many microservices running on top of a Kubernetes cluster.
- Created a mobile application that can analyze positioning, pricing, and out-of-stock goods, relying only on the photo of the product, and works with all types of packaging and product placement called Shelf Inspector.
Senior Full-stack Machine Learning Engineer | CTO
Expansion of Eternity AG
- Developed an end-to-end solution to simulate a treatment's facial aesthetical impact.
- Trained Mask R-CNN models to detect possible treatment areas and advise on areas where doctors should focus.
- Developed a complete back-end solution consisting of REST API for a front-end application and a payment gateway, using a PostgreSQL database for storage.
- Implemented a precise detection of the face on the taken image and accurate detection of facial features like nose, eyes, mouth, etc.
- Detected the location of points on the face with very high accuracy for the injection of a substance for aesthetical treatment.
Senior Full-stack Machine Learning Engineer | Co-founder
- Developed computer vision algorithms to detect defectively manufactured parts with high accuracy. It requires precise object detection and recognition algorithms that distinguish the most minor object characteristics.
- Implemented the notification of safety hazards. An area with robots requires specific safety rules: wearing a hard hat, glasses, and sometimes gloves. Implemented a solution to check all safety rules are followed from the feeds of security cameras.
- Created algorithms to navigate industrial robots using the robot operating system from a video feed. Detected obstacles and other objects to avoid robots crashing.
- Designed the whole product architecture from scratch as CTO of the company.
Head of Development, Search Engine
- Redesigned the architecture and generated technical solutions for our full-text core technology, which enabled the widespread adoption of our solution across the whole company in multiple projects.
- Spearheaded a modernization of the development process by establishing the brand-new roles of DevOps and SRE, which resulted in a faster process of releasing the changes into production.
- Delivered new functionalities such as an image search engine which was astoundingly successful in the highly competitive market of search engines in Central Europe.
Senior Back-end Developer
- Developed an internal system for objective quality measurement of the full-text search system. It allowed product managers to focus on important features with the most benefit for the end-users.
- Scaled the core of the full-text engine to processing a triple amount of data.
- Redesigned the architecture of the system for taking screenshots of internet pages on a scale of thousands of pages per second.
In a small team of three members, I was the lead machine learning engineer and back-end developer in the build of this app. Due to the nature of the application, we paid particular focus on generating trustful results while maintaining high-security standards.
Seznam.cz Full-text Search Enginehttps://search.seznam.cz/
I started this project at the beginning of my career, and I participated in developing this search engine as a back-end developer within a large team. A few years later, I became head of the development of the whole full-text platform. This was when I became increasingly passionate about algorithms driven by data and machine learning, and after that, I decided to start my own business.
Robotic System Air Hockey Gamehttps://fb.watch/6kAEW9qBGv/
I was a lead machine learning engineer, and I was responsible for creating the algorithms from the proof of concept to production-ready systems.
To demonstrate our ability, we created a robotic system that can play an air hockey game against a human player. The game was publicly announced in a trade fair called Amper in 2019. You can still see the video at the link below.
It was a real-time computer vision system with multiple object detection while. being able to control and navigate with the robotic 6-axis arm.
Extracting Information from Invoices
I worked on this project as the lead machine learning engineer and responsible for designing solutions from scratch to production.
The solution is divided into three major stages. In the first stage, we standardized the format of incoming invoices. It includes some computer vision tasks and classification models based on convolution neural networks. In the next stage, we implemented optical character recognition for extracting raw text from invoices. The last step consists of named entity recognition and some post-processing for the final classification of important information on the invoice.
We also used production algorithms in highly scalable microservices running in a Kubernetes cluster that communicated over a RabbitMQ message broker.
Flask, Bootstrap, Spark, Svelte, Hadoop
Fast.ai, PyTorch Lightning, REST APIs, Keras, PyTorch, TensorFlow, Scikit-learn, OpenCV, SQLAlchemy, React, Pandas, NumPy, Node.js, Flask-RESTful, XGBoost, PySpark, Spark ML, Natural Language Toolkit (NLTK), Azure Cognitive Services, SpaCy, Google Vision API
Docker Compose, Git, RabbitMQ, NGINX, GitHub, Scikit-image, Grafana, Pytest, Helm, Ansible, Docker Swarm, Expo, TensorBoard, Azure Kubernetes Service (AKS), Named-entity Recognition (NER), GitLab CI/CD, Azure Machine Learning, Google Cloud AI, Amazon SageMaker
Data Science, REST, Continuous Integration (CI), Test-driven Development (TDD), Continuous Development (CD), Agile, Scrum, Kanban, Microservices, DevOps, Continuous Delivery (CD), Management, ETL
Linux, Docker, Jupyter Notebook, Azure, AWS Lambda, Kubernetes, Databricks, Amazon Web Services (AWS), Google Cloud Platform (GCP), Apache Kafka
Amazon S3 (AWS S3), PostgreSQL, MongoDB, MySQL, Redis, Elasticsearch, Memcached, Databases, Database Modeling, Data Pipelines, Google Cloud, Azure Cosmos DB
Artificial Intelligence (AI), Computer Vision, Neural Networks, Machine Learning, Data Engineering, Cloud, Full-stack, Machine Learning Operations (MLOps), IceVision, Software Architecture, Back-end, FastAPI, OCR, Computer Vision Algorithms, Convolutional Neural Networks, Sanic Web Server, Deep Learning, Image Recognition, Facial Tracking, Containers, Containerization, CI/CD Pipelines, APIs, Machine Vision, AI Design, Build Pipelines, GitHub Actions, Object Detection, Image Processing, Tesseract, Pytesseract, Deep Neural Networks, Image Analysis, Image Annotation, Data Extraction, Architecture, Real-time Data, Amazon Textract, Leadership, Big Data, Prometheus, Generative Adversarial Networks (GANs), Natural Language Processing (NLP), Web Security, IT Project Management, IT Product Management, Web App Security, Data Visualization, Facial Recognition, Data Modeling, Predictive Modeling, Predictive Analytics, Performance Tuning, Performance Testing, Data Analysis, Data Analytics, Algorithms, Neptune, MLflow, Weights & Biases, Metabase, Open Neural Network Exchange (ONNX), Biometrics, Health, Amazon Machine Learning, Google Cloud Machine Learning, ML Pipelines, Object Tracking, DocTR, Web Development, Hardware, CCTV, ChatGPT, Generative Pre-trained Transformer 3 (GPT-3), Vertex, Vertex AI, Hugging Face, GPT, Generative Pre-trained Transformers (GPT), DVC, Data Versioning, MLEM, Handwriting Recognition, Robot Operating System (ROS), GPU Computing, Robotics, ABB Robots, Metaflow
Bachelor's Degree in Information Technology
Brno University of Technology Faculty of Information Technology - Brno, Czech Republic