HAVA: Maritime Criminal Detection
NLP pipeline for detecting illegal fishing and maritime criminal behaviour, built with Clause Technology and NCIS
While consulting with Clause Technology, in collaboration with NCIS, I built an automated system for detecting illegal fishing activity and maritime criminal behaviour from open-source reporting.
- Automated ingestion: web scraping of incident reports with NLP-based classification
- Entity recognition: BERT/NER pipelines for vessels, locations, and organizations, with LLM-assisted similarity de-duplication
- Deployment: dockerized architecture on MongoDB/PostgreSQL
On the same engagement I also developed a PDF table parsing pipeline (computer vision and OCR for complex table structures, rotated tables, multi-table pages, with image preprocessing for OCR and variable extraction/standardization).
The project builds on a capstone project on vessel trajectory analysis that I supervised with NCIS. The methods carry over directly from my clinical work: multi-source data engineering, class-imbalanced prediction, and human-in-the-loop triage.