Enhancing environmental crime investigations with AI
KickstartAI x Politie Nederland
Environmental crime is surging globally, yet complex regulations often hinder prosecution. While offenders may evade enforcement within environmental law itself, they remain vulnerable through related domains such as fraud, money laundering, and transport law. In partnership with the Dutch Police (High Impact Environmental Crime team), we are developing an AI-powered intelligence assistant to help investigators identify these legal connections and uncover offenses that might otherwise remain hidden.
Uncovering hidden violations in complex cases
Manually cross-referencing legislation, case law, and explanatory notes remains a significant bottleneck for intelligence officers. At the same time, because environmental cases often span multiple legal domains, relevant violations can be difficult to identify – particularly for investigators who are not specialists in every area involved. As a result, important leads may be overlooked, slowing the transition from analysis to active investigation and limiting prosecution opportunities.
Goal
The goal of the project is to develop an AI solution that helps identify relevant laws and potential violations across domains, accelerating the research phase to enable faster investigations. By streamlining investigative work and surfacing hidden connections, the tool can boost prosecution efforts and contribute to protecting the environment and public health.
Approach
The approach consists of three phases:
Use Case Identification: Deeply understanding the current process and defining the specific problem to be solved.
Model Development: Building the tech infrastructure and training the core AI models.
Implementation and Adoption: Integrating the solution into the workflows of the Police.
Key features
Our solution enriches initial investigation signals with automated legal context. By processing unstructured research documents, the tool performs cross-domain correlations against a vast library of open-source data. This surfaces a prioritized list of regulations, ensuring officers evaluate legal angles beyond their primary expertise.
The system utilizes a Retrieval-Augmented Generation (RAG) pipeline to structure data into a searchable vector database. A Large Language Model (LLM) then employs hybrid search and reranking to deliver precise, context-aware results. Specifically, it will enable:
Cross-domain correlation: Automatically identifying violations across environmental, fraud, transport, and anti-money laundering laws.
Prioritized legal suggestions: Providing a ranked list of relevant laws and jurisprudence based on textual input from investigation documents.
Public data integration: Scraping and analyzing data from Wetten.nl, Rechtspraak.nl, and other specialized legal journals.
Expected Results
While the project is currently in the development phase, it is designed to achieve the following:
Faster investigation throughput: Reducing the manual effort required to find non-obvious legal links.
Higher case quality: Ensuring comprehensive and legally sound evidence packages
Enhanced expertise: Enabling intelligence officers to proactively identify violations and legal angles outside their primary specialization.
Scalable AI adoption: Validating the core engine’s ability to perform accurate correlations to ensure long-term user trust and adoption.