Dimitrios Dagdilelis

AI/ML Research Engineer

Currently

I’m an AI engineer, passionate about transforming data into knowledge. Empowering ML teams by advocating MLOps best practices, developing better data tools and making models transparent. I design and deliver data products from concepts to production, keeping myself close to the data and involved in the entire ML product lifecycle. Five years of experience on using AI to develop autopilot systems and geospatial data analysis applications.

Specialized in

• Building scalable AI systems
• Computer vision, 3D perception, and geospatial data
• Multimodal Data Fusion

Professional Experience

2020 - 2023 __AI/ML Research Engineer , ShippingLab, Copenhagen, Denmark

• Led the development of a modular perception stack of an autopilot system, successfully delivering the data product despite disruptions caused by the pandemic (preview-link).
• Designed a transformer-based multi-modal fusion architecture for 3D scene understanding and target tracking, integrating camera, LiDAR, and radar inputs to enhance spatial perception and semantic reasoning. (preview-link).
• Developed the ML perception stack for Denmark’s first autonomous ferry operation (preview-link).
• Integrated automatic and semi-automatic data annotation processes, reducing manual labeling costs (preview-link).
• In communication with downstream product owners, identify edge cases where ML is failing, pinpointed training data availability issues, then design and execute real, as well as synthetically generated data collection campaigns that solve the performance issues.

2024 - Present Senior AI & MLOps Enginer, National Danish Security Organization, Copenhagen, Denmark

• Top-secret security cleared, authorized to handle classified information
• Designed and put in production a computer vision agent, responsible for real-time object recognition in high-volume geospatial image data.
• Implemented a visual search system by learning deep visual embeddings and enabling image-based retrieval across large-scale indexed datasets.
• Contributed to retrieval-augmented generation (RAG) workflows by integrating image and metadata retrieval pipelines with downstream language models.
• Developed an auto-labeling framework, self-supervised learning, active learning, and domain adaptation to minimize label cost and maximize their utilization.
• Built explainable-AI features into the image processing API, aiding interpretability and supporting human-in-the-loop validation.
• Designed and implemented end-to-end MLOps pipelines for automated data preprocessing, continuous model training & versioning, and drift monitoring in production.
• Responsible for developing model evaluation tooling, as well as enhancing the robustness, generalization, and performance of deployed vision models.
• Fostered internal knowledge-sharing by hosting weekly ML reading groups and cross-functional workshops.
• Work specifics under confidentiality; details provided upon request in a personal interview.

2022 Visiting AI Engineer, SeaAI, Vienna, Austria

• Proposed and implemented a solution to extend the product’s 2D object detection capabilities into 3D object detection.
• Collaborated remotely, asynchronously as well as in person with the AI team in Vienna.
• Curated datasets, and developed data visualization tools across a large-scale NoSQL database.

Education

2024 PhD in Cyber-Resilient Data Fusion, Technical University of Denmark

• Focus on multimodal data fusion, deep learning, and 3D computer vision. If curious, find my thesis here.

2020 MSc in Automation & Robot Technology, Technical University of Denmark

2016 Diploma in Electrical & Computer Engineering, Aristotle University of Thessaloniki

Technical Skills

• Languages: Python, SQL, Bash
• ML Frameworks: PyTorch, TensorFlow, Scikit-learn, HuggingFace Transformers
• Cloud: AWS, Google Cloud
• MLOps: Docker, MLFlow, FastAPI, ClearML, Weights & Biases, Grafana, Airflow
• Tooling: FiftyOne, OpenCV, GDAL, Label Studio, CVAT, DVC, NumPy, Pandas
• Data: PostgreSQL, MongoDB, HBase
• Deployment & Serving: FastAPI, TorchServe, ONNX
• Monitoring & Logging: Prometheus, Grafana
• Experimentation: Weights & Biases, MLFlow
• Modalities: images, video, text, point-clouds, radar, semantic maps
• Collaboration: GitHub, GitLab, JIRA, Confluence, Slack