Andrei-Alexandru – Python, Snowflake, PyTorch

Industry leading analytics SaaS in Europe. Andrei has built the engineering team.

• Andrei has been working on a custom OCR engine, a custom detector, an insanely fast and compute-light forecasting engine (statistical and neural forecasting).
• Orchestrating ML workflows using Prefect.
• Conducting ML Research (YOLO and leading OCR models).
• Data engineering (Prefect, MySQL, DynamoDB, Mongo, SES, AWS Batch, AWS Athena, AWS Glue, AWS RDS, Azure Storage, GCP Storage, Python, pandas, polars, deltalake, sql) and Data analysis(Microsoft PowerBI, Tableau, Excel, Sheets).
• Responsible for data integrity inside the StockRx platform: single source of truth for medication/pharmacy products.
• Leading web development (SOLID, REST APIs, .Net, React, Recoil, Typescript/Javascript, AWS SQS, AWS Lambda, Dynamo DB, AWS API Gateway, observability (honeycomb) etc).
• Setting up DevOps (AWS, GCP, Docker, Gitlab workflows, Github actions, SAM, IaC(terraform, cdk, cloudformation), billing, architecture.

Research project to benchmark OCR models and trends. Evaluated models and benchmark price per page and supported mime types.

• Led a research project focused on benchmarking state-of-the-art Optical Character Recognition (OCR) models and market solutions as of 2023.
• Conducted a systematic evaluation of open-source and commercial OCR engines (e.g., Tesseract, EasyOCR, AWS Textract, Azure Form Recognizer, Google Document AI).
• Designed benchmarking criteria including recognition accuracy, processing speed, supported MIME types (PDF, TIFF, PNG, DOCX), API capabilities, and real-world robustness.
• Calculated and compared price-per-page metrics across providers to assess cost-efficiency for large-scale document processing pipelines.
• Built automated test pipelines in Python to batch-process large datasets, collect model outputs, and generate comparative performance reports.
• Analyzed OCR model trends (e.g., transformer-based OCR models, document layout analysis, handwriting recognition advances) and provided strategic recommendations for model adoption based on client needs and project constraints.
• Delivered a comprehensive report summarizing technical findings, performance metrics, cost analysis, and future-readiness of OCR solutions.

A project with the primary goal of delivering insights to the sales team about accounts. The machine learning model served predictions for each account, and then the resulting table was uploaded into Snowflake, connecting to a PowerBI dashboard. Everything was scheduled for a weekly run with the process running fully automated (tests included) at the end of the project.

• Led the development of a cloud-based data processing and predictive modeling pipeline to deliver business insights to the sales team (account healthscore).
• Designed and implemented feature engineering workflows and machine learning models using Python and SQL, leveraging AWS services (EC2, Batch, ECS) and Snowflake as the data warehouse.
• Built automated scripts packaged in Docker containers and orchestrated through AWS Batch on ECS for scheduled weekly runs.
• Developed custom ETL processes (Airflow, dbt, python, seekwell, SQL, fivetran) to extract, transform, and load relevant datasets from Snowflake, performed advanced feature engineering, trained models offline, and generated actionable insights.
• Handled full automation and deployment, including configuring compute resources on AWS EC2 instances, monitoring job execution, and maintaining versioned deployments.
• Delivered consistent, high-quality outputs to business stakeholders, supporting data-driven sales strategies.

This research aims to make an experimental study on a standalone system to evaluate how various hardware configurations affect the overall performance of deep learning.

• Optimized running deep learning models with Fourier transform;
• Optimized running deep NNs by performing lots of INT quantizations and lots of literature research;
• Ran the slide algorithm and analyzed it with the benchmark;
• Models analysis, pruning and ablation studies
• Modified how deep learning models run on CPUs;
• Analyzed the research results and showed that the performance greatly relies on the hardware configurations.
• Integrations with MatLab
• Process engineering
• Lots of python and pytorch
• Research paper written in Latex, rejected twice from ICML and NeurIPS

Designed and led the development of a Visual Inspection Automation (VIA) system for real-time ceramic plate defect detection at one of Europe's largest ceramic manufacturers. Built a YOLO-based computer vision pipeline for detecting cracks, chips, and surface defects, achieving high precision and real-time inference speed. Led integration with industrial Fanuc robots for automated handling and rejection of defective plates, ensuring seamless operation within the production line. Managed data acquisition (industrial cameras), dataset creation, model training, optimization (precision, recall, mAP), and deployment under strict production constraints. Delivered a robust system that significantly improved defect detection rates and reduced manual inspection costs, contributing directly to operational efficiency gains.

• Created a deep learning model + industrial infrastructure that passed the pilot line.
• Tested and customized an object detection model.
• Created a robust data engine (version control with DVC & hosted on Azure).
• Benchmarking of computer vision models, pytorch implementations.
• Created an active learning pipeline Integration with production servers and Siemens PLCs.
• Started development for the Nvidia Jetson Xavier board.
• Tried defect tracking in order to get rid of sensor integrations.
• Was in change of product management and bulding a team of contractors and employees.

Working as a computer vision engineer in an industrial setting. Optimization of computer vision activities. Data science. Maintenance.

• Developed and optimized computer vision models for defect detection, quality control, and visual inspection tasks in industrial production environments.
• Applied deep learning techniques (CNNs, object detection, segmentation) and classical vision methods (OpenCV pipelines) to meet specific production requirements.
• Performed model optimization for inference speed, resource constraints, and deployment on edge devices or industrial servers.
• Conducted continuous maintenance for industrial hardware and software.
• Built data preprocessing pipelines for image cleaning, augmentation, annotation consistency, and labeling management.
• Worked closely with cross-functional engineering teams (mechanical, automation, software) to align technical solutions with production workflows and business objectives.

Car license plate recognition is real-time detection and recognition of the license plates. The machine learning model of which is used for a smart parking solution. Used CRNN CTC, custom LSTMs and YOLO models. Also maintained a 3d labeling solution and benchmarked multiple labeling solutions. Versioning with dvc and git.

• Created a CRNN-CTC model that classified and detected license plate numbers/letters from 80x26p images.
• Worked on car tracking algorithms with ScaledYOLOV4, DeepSORT and the clip model with zero-shot oriented tracking.
• Maintained a custom labeling solution and benchmarked off the shelf ones.
• Worked on custom LSTM for license plate recognition.

Simple document management platform for a SME client hosted on the cloud.

• Designed and implemented backend services using .NET Core for secure document upload, storage, retrieval.
• Developed a React-based frontend for user-friendly document management (upload, search, access control).
• Architected and deployed the full solution on AWS, leveraging services like S3 (file storage), RDS (database), Cognito (authentication), and EC2/ECS (application hosting).
• Containerized the application using Docker and orchestrated deployment through AWS ECS/AWS Lambda.
• Designed and optimized relational database schema (AWS RDS) for efficient document metadata storage and querying. Dynamo DB for NoSQL document storage.
• Implemented CI/CD pipelines
• Ensured high availability, data durability, and security by applying AWS best practices (IAM policies, encrypted storage, HTTPS endpoints).
• Conducted performance tuning and cost optimization of AWS infrastructure to match SME client requirements.
• Provided technical documentation and support for client onboarding and system administration.