Maksym

Maksym Demkiv
[відкрити контакти](див. вище в блоці «контактна інформація») | [відкрити контакти](див. вище в блоці «контактна інформація») | [відкрити контакти](див. вище в блоці «контактна інформація») | github.com/lDereQl

About me
AI Backend Engineer focused on building production-grade systems at the intersection of LLMs, backend architecture,
and applied machine learning. Experienced in designing on-premise AI agent infrastructures, developing deterministic
backend logic around probabilistic models, and integrating embedded systems for real-time applications. Strong
background in Python, system design, and model evaluation, with a practical, engineering-first approach to scalable and
reliable AI solutions.

Experience
AI Backend Engineer Intern October 2025 - January 2026
Ingram Technologies Brusseles, Belgium
• Improved reliability of an on-premise AI agent platform by building production-grade FastAPI services with

automated testing and CI integration, increasing overall system stability.
• Enhanced agent performance and response consistency by designing structured prompt templates and session-based

memory management, resulting in more predictable LLM outputs.
• Strengthened deployment and observability by optimizing Docker Compose configurations and CI/CD pipelines

with automated linting, logging, and health metrics, improving production readiness.
Software Engineer March 2025 - September 2025
Baltic Viper Kaunas, Lithuania
• Improved real-time drone control capabilities by integrating microcontroller-based components, resulting in stable

telemetry processing and reduced signal latency.
• Accelerated development of an AI-powered target lock and guidance system by leading a 5-member engineering

team, resulting in a functional FPV drone prototype.
• Increased team delivery predictability by applying Scrum methodology to sprint planning and scope management,

resulting in more consistent milestone completion.
AI/LLM Engineering Intern January 2025 - March 2025
Hyand Kaunas, Lithuania
• Optimized AI-assisted development workflows by benchmarking multiple LLMs using vLLM and Ollama, resulting

in data-driven model selection decisions.
• Improved maintainability of a shipping application by designing backend logic using Domain-Driven Design

principles, resulting in clearer domain boundaries and scalable architecture.
• Reduced production risks by evaluating performance and security implications of AI agent deployment, resulting in

safer integration into existing workflows.

Projects
ECG Signal Quality Annotation – Bachelor’s Thesis February 2026 – Present
• Technologies: Python, PyTorch, scikit-learn, NumPy, Pandas, WFDB, pytest

• Developing an ML framework for automated segment-level quality annotation of long-term wearable ECG

recordings, separating deterministic signal preprocessing from model inference to ensure reproducibility.
• Implemented a 1D CNN-LSTM classifier trained on the BUT QDB dataset (24h+ free-living recordings) with a

3-class quality scheme aligned to expert sample-level annotations.
• Established SVM and Random Forest baselines using hand-crafted signal quality indices (SQIs) for comparative

evaluation across MIT-BIH and PhysioNet datasets.

DMAI – AI Dungeon Master Platform February 2026 – Present
• Technologies: Python, PostgreSQL, SQLAlchemy, Alembic, Pydantic, Docker, Ollama, pytest

• Built a text-based RPG engine implementing D&D 5e combat mechanics by separating deterministic game logic

from LLM narrative generation to prevent state hallucination.
• Achieved stable session persistence by implementing optimistic locking and checkpoint/rollback support on top of

PostgreSQL with Alembic migrations.
• Maintained 90%+ test coverage across domain and engine layers by writing unit and integration tests with pytest.
 Sperm Cell Tracking & Analysis System September 2025 – December 2025
• Technologies: Python, PyTorch, OpenCV, YOLOv8n

• Built an automated sperm cell detection and tracking system by preprocessing 20 microscopy videos, resulting in a

clean, validated dataset for model training.
• Trained a YOLOv8n detection model on the prepared data, achieving 98.9% [email protected] and 80.5% [email protected]:0.95

under limited GPU resources.
• Implemented real-time multi-object tracking with OpenCV, processing 3,000+ frames at ∼40 FPS and tracking

1,700+ unique cells with stable trajectories.

Technical Skills
Programming Languages: Python, C/C++, SQL (PostgreSQL)
Backend & Architecture: FastAPI, Domain-Driven Design (DDD), REST APIs, Microservices, CI/CD Pipelines
AI / LLM Engineering: PyTorch, scikit-learn, CNN, LSTM, YOLOv8, OpenCV, LLM prompt engineering, Session
memory design, vLLM, Ollama
Data & ML Tools: NumPy, Pandas, WFDB, Model benchmarking, Signal preprocessing, Feature engineering (SQIs)
Databases: PostgreSQL, SQLAlchemy, Alembic, Transaction management, Optimistic locking
DevOps & Tooling: Docker, Docker Compose, GitHub Actions, Automated testing (pytest), Logging & observability,
Git
Embedded Systems: Microcontroller integration, Real-time telemetry systems

General Skills
Softskills: Leadership, Communication, Teamwork, Analytics, Agile Project Management
Languages: English (C1), Ukrainian (Fluent), Russian (Fluent), Lithuanian (A1)

Education
Kaunas University of Technologies Kaunas, Lithuania
Bachelor of Arts in Computer Science – Artificial Intelligence Expected June 2026
Lviv Physics and Mathematics Lyceum Lviv, Ukraine
Aug. 2017 – May 2021