Oleksandr

[открыть контакты](см. выше в блоке «контактная информация») Portfolio:
Bratislava, Slovakia Oleksandr Vahabov github.com/
[открыть контакты](см. выше в блоке «контактная информация») AI / ML Engineer [открыть контакты](см. выше в блоке «контактная информация»)

Engineer specializing in the design and deployment of intelligent, production-ready systems. Experienced in building multi-agent
AI architectures using LangGraph and LangChain, fine-tuning transformer models for NLP tasks, and integrating LLM APIs into
automated end-to-end pipelines. Proven ability to engineer high-performance FastAPI backends for real-time model serving,
orchestrate async microservices with RabbitMQ, and containerize complex AI ecosystems with Docker. Skilled in computer vision,
anomaly detection, and scalable data engineering with PySpark and PostgreSQL.
SKILLS
Technical Stack Python, Java, LangGraph, LLM, SQL, NoSQL, REST API, FastAPI, Spring Boot, PySpark, Docker,
Kubernetes, Cloud Computing Services (AWS, Azure).

Communication English (C1), Slovak (B2), German (B1), Ukrainian & Russian (Fluent).

TECHNICAL EXPERIENCE
Data Developer 03/2026 — PRESENT
Adastra Bratislava, Slovakia
• Developed and maintained ETL pipelines, data processing scripts, and transformation workflows to deliver clean, accurate
datasets for reporting and decision-making.
• Contributed to external client projects through technical consultation, solution design, and cross-functional collaboration with
data engineers and business stakeholders.
• Provisioned a self-hosted GitHub Actions runner on a Linux VM, automating setup via Bash scripts and containerizing the
environment with Docker.
• Result: Delivered a cost-effective, self-managed CI/CD infrastructure giving the company full control over build resources and
runtime dependencies.

AI Engineer 03/2025 — 03/2026
TeamChallange Bratislava, Slovakia
Project 1: PII Removal System
• Developed a PII anonymization pipeline using fine-tuned XLM-RoBERTa NER models and rule-based heuristics.
• Built a high-performance web interface and API using FastAPI to serve the model and handle real-time document redaction.
• Engineered data ingestion from Azure Blob Storage with OCR support for automated batch document processing.
• Integrated OpenAI API for secondary validation of model predictions in complex unstructured data.
• Containerized inference services via Docker to ensure consistent deployment across cloud environments.
• Result: Achieved 95%+ F1 score for NER tasks and scaled throughput to 500 documents/minute.
Project 2: Client Fetcher
• Architected a multi-agent system using LangGraph with two agents: Classifying one for incoming messages and storing leads
into database, and another for retrieving leads and autonomously conducting a dialogue.
• Leveraged prompt engineering techniques to ensure consistent and accurate classification.
• Integrated the multi-agent AI layer with an async PostgreSQL and Telegram API (Telethon) into a containerized Docker pipeline.
• Result: Delivered an autonomous multi-agent AI system that handles lead detection, qualification, and escalation, replacing
manual lawyer outreach.
Project 3: Backend-Shop
• Built a microservices architecture using Spring Boot with four services: API Gateway, Auth, Product, and Notification.
• Implemented JWT authentication, role-based route protection, Google OAuth2, and password reset via email.
• Used RabbitMQ for async messaging between services and Cloudinary for product image storage.
• Added dynamic product filtering and sorting with JPA Specifications and email subscriptions with Thymeleaf templates.
• Containerized all services with Docker for consistent deployment.

PROJECTS
Real-time Fraud Detection System (In Active Development) MLOps / Data Engineering
• Implementing a high-frequency data ingestion layer simulating transaction streams stored in LocalStack (AWS S3).
• Developing an anomaly detection model using PyTorch Autoencoders.
• Building an ETL pipeline in PySpark to enrich real-time transaction data with historical user profiles from PostgreSQL.
• Orchestrating the infrastructure using Docker and Kubernetes.
• Status: currently optimizing the Spark-to-Inference workflow for low-latency processing.
LabelMe Object Recognition System Computer Vision
• Developed a multi-label classification pipeline for the LabelMe-12-50k dataset
• Addressed extreme class imbalance by implementing Data Augmentation techniques to improve minority class representation.
• Optimized a ResNet-based architecture with skip connections.
• Result: Achieved 91% accuracy in multi-class tasks and 80.2% subset accuracy in multi-label detection, with F1-scores
exceeding 0.85 for key categories.
EDUCATION
2024 – 2027 Bachelor of Science: Computer Science, Slovak University of Technology, Bratislava, SK.