Each year, the Norwegian public spends about 5.6 billion NOK on planning and validating building applications. The process is complex, time-consuming, and often yields poor-quality submissions. The KartAI project aims to streamline this by developing AI tools to support the application process. The project is a collaboration between the Norwegian University of Science and Technology (NTNU) and KartAI.
KatAI's main goal is to streamline municipal work processes related to cadastre and building case processing using automated, advanced data-driven methods, including artificial intelligence in combination with proactive user and citizen dialogue. The goal is to contribute to automating and streamlining the processing of building cases.
The scope of the project consists of two primary objectives:
-
Develop a Web Application: This platform will serve as a centralized hub, integrating the various AI models available through the KartAI project. By bringing these models together, the application will serve as a proof-of-concept (PoC), allowing KartAI to display their assortment of AI tools for different stages of the application process.
-
Create a Summary AI Assistant: This AI-driven tool analyze documents from submitted applications and generate concise summaries, highlighting key points. The system implements a checklist matching feature. It cross-reference the building application with an official checklist and relevant regulations and inform about the quality of the application. This functionality is designed to support both applicants and case workers, enhancing the overall efficiency and clarity of the application process
The following is a brief overview of the final product, including the system architecture and screenshots of the web application and the AI Summary Assistant.
Click to see Web Application screenshots
-
Landing Page
The landing page for the web application.
-
Landing Page with Navbar
The landing page with the navigation bar displayed.
-
3D Tiltaksvisning
The page for 3D visualization of projects.
-
PlanChat
A chat window designed to answer questions about laws and regulations.
-
ArkivGPT Interface
The user interface for interacting with the ArkivGPT AI model.
-
ArkivGPT Results
Displaying results from an ArkivGPT query.
-
File Preview
Previewing files related to ArkivGPT queries.
-
CADAiD Request
User interface for requesting validation from the CADAiD model.
-
CADAiD Results
Results generated from the CADAiD model.
-
Applications Overview
A page showing an overview of applications for municipality workers.
-
Municipality Dashboard (Top)
The dashboard for municipality workers showing checklist maps.
-
Municipality Dashboard (Bottom)
The dashboard displaying checklist maps and AI model results.
-
User Dashboard
The dashboard where applicants can review their applications using various AI models.
Click to see AI Summary Assistant Screenshots
- AI Summary Assistant
The graph showing the AI agent structure.
- LangSmith tracking
The monitoring of the agent showing what choices it makes.
Here one can see the agent have retrieved relevant laws and regulations from vector database and done a web search, as well reflect on the output of it self before marking the checkpoint and giving its reasoning.
Before you start, make sure the following tools are installed on your system:
- Git: Version control system to clone the project repository Download Git
- Docker: To containerize the application and ensure it runs consistently across different environments Download Docker
Start by going into the /webapp folder, making a copy of the .env.example file and renaming it to .env. This file contains the environment variables that the application needs to run. Open the .env file and update the environment variables according to your local or production setup.
To run the project, you can use the following commands:
docker compose --env-file ./webapp/.env --env-file ./backend/.env up --build -dThis command will build the Docker images (if necessary) and run the containers in the background. You can access the clientside code at http://localhost:3000 and the API at http://localhost:8000. The Swagger documentation for the API is available at http://localhost:8000/docs.
To stop the containers, you can use the following command:
docker compose downImportant: In order to achieve the full functionality of the application, the AI models from the KartAI project must also be running. In our development we have ran them as docker containers locally on our machines. Though in the future, these will hopefully be available as public APIs.
The team behind this project is a group of students at NTNU in Trondheim. The team consists of:
 Andreas Lilleby Hjulstad |
 Artemis Kjøllmoen Aarø |
 Johanne Eide Omland |
 Magnus Giverin |
 Maurice Wegerif |
 Sverre Nystad |