AI streamlines statutory document discovery, saving 1000’s of hours so DoC staff can get back to conservation.
Problem
The Department of Conservation (DoC) manages a large and complex set of statutory documents that underpin concession approvals, regulatory enforcement, and operational decision-making. These documents are often hundreds of pages long, and many are scanned PDFs — making even basic keyword searches difficult or impossible. While teams across DoC have worked to manage this information manually, reviewing and interpreting these documents remains time-consuming and onerous, especially when under pressure to answer queries from rangers or external stakeholders. This administrative load diverts skilled staff away from their core mission: protecting New Zealand’s unique species and environments.
DoC had trialled multiple off-the-shelf RAG and vector search tools, but found that none could deliver accurate or contextually relevant answers — especially for queries involving complex geographic hierarchies, such as activities governed by overlapping conservation management plans, bylaws, and land features that span multiple parks or regions. What was needed was a solution that could reason through geospatial relationships and legal context, not just retrieve text.
Solution
In four weeks we delivered a secure, portable AI assistant that allows DoC staff to ask natural language questions about conservation law, statutory areas, and permitted activities — and receive instant, citation-backed responses. The assistant combines Retrieval-Augmented Generation (RAG) with a Neo4j knowledge graph, ensuring answers are grounded in the correct context and traceable to source documents.
Each response includes direct links and page-level citations to the relevant statutory document allowing staff to immediately open the original PDF and verify the source. This makes it easy to quote or reference exact legislation when drafting official communications, responses to concession applications, or internal policy documents. It also de-risks hallucinations, as staff can quickly validate that the AI’s output is rooted in official content.
The system extracts structured data from over 140 statutory documents, builds relationships between areas, activities, and legal constraints, and delivers results via a secure chat interface. Key technologies include LangGraph, Neo4j, Azure OpenAI, and Azure Container Apps, with all infrastructure provisioned through Terraform for easy handover to DoC’s Azure environment.
Results
The AI assistant transforms how statutory content is accessed, cutting research time from days or weeks down to hours. Staff can now get clear, grounded answers with source citations — whether they’re reviewing a concession request, supporting rangers, or preparing internal guidance.
By removing the friction from document discovery, DoC staff are empowered to spend more time in the field and less time in PDFs — directly contributing to the department’s conservation outcomes. The architecture is modular and production-ready, with future phases scoped to extend coverage and integrate with broader internal systems.
Technologies
