Building an End-to-End Automated AI Data Pipeline for Maritime Safety
Maritime Safety & Security AI Systems
Video-based Object Detection, Classification, Anomaly Detection
Real-time Surveillance Video and Image Data
Captain’s Eye was founded in 2020 by Capt. Uri Ben Dor and Doron Oizerovich. The company developed a holistic AI-based system that detects unusual events in real time, helping prevent property, physical, and financial damage that might occur at sea or inland. The system can identify and alert operators to safety and security issues in all types of vessels, such as smoke, leakages, security breaches, unsafe crew behavior, or anomalies. Its solutions serve the maritime industry and extend to ports, plants, and other high-risk sectors.
Dataloop is an enterprise-grade data infrastructure platform for AI, designed specifically for unstructured, multimodal data. It provides the data-centric foundation required to support the entire AI lifecycle – with powerful tools for unstructured data management and AI orchestration pipelines, it delivers the AI-ready Data Stack, empowering leading enterprises to build, deploy, and scale AI solutions with confidence and speed for the next wave of AI.
As organizations expand their AI initiatives, the volume and complexity of data operations naturally increase. Real-time video and image streams at scale demand workflows that are automated, reliable, and transparent. For Captain’s Eye, delivering accurate anomaly detection across fleets required continuously preparing large volumes of video data for training and validating high-performing AI models. To achieve this, the company set out to unify its unstructured data lifecycle within a fully orchestrated AI workflow. The goal was to maximize automation, ensure high data quality and provide greater visibility and control across the data pipeline.
Captain’s Eye integrated Dataloop into its process to build a fully orchestrated workflow that transforms unstructured video streams into AI-ready datasets. Real-time feeds from vessels are automatically ingested and enriched with metadata such as vessel ID, fleet, and incident type — a step that ensures data is structured, consistent, and ready for downstream analysis.
From ingestion onward, the workflow runs as a seamless, fully orchestrated process that transforms raw video streams into structured, AI-ready datasets. Automated enrichment, validation, and versioning ensure data consistency and balance, while orchestration pipelines streamline preparation for training, validation, and testing cycles. This unified approach accelerates data readiness and supports scalable, high-performing AI development.
Illustration of how Port Safety Monitoring data can be clustered on the Dataloop platform, demonstrating how customer datasets can be visualized and organized for safety and anomaly detection use cases
Screenshot of an Active Learning Pipeline template on the Dataloop platform – designed to automate and streamline iterative model training for unstructured data at scale
The result is a highly automated process that maximizes speed while enhancing quality and control. Captain’s Eye now prepares datasets 30% faster and supports seven AI pipelines in parallel. AI data engineers manage workflows through Dataloop’s interface without heavy coding, while developers extend functionality with custom code nodes for specialized maritime use cases. Together with Dataloop, Captain’s Eye strengthened its ability to accelerate data preparation, scale operations, and drive AI innovation across fleets.
Time to AI-ready dataset
30% faster through automation
Data quality assurance
95% automated through smart sampling
Use cases supported in parallel
7% AI pipelines managed concurrently
AI Team Lead
“With Dataloop, we built a workflow that makes data preparation fast, structured, and reliable. We now prepare datasets 30% faster and support multiple models in parallel — giving us the ability to scale AI across fleets with confidence.”
