Developing Advanced Northern Lights AI-Assistant with OpenAI API
Vanhatapio, Juhani (2025)
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025120833415
https://urn.fi/URN:NBN:fi:amk-2025120833415
Tiivistelmä
This development-oriented thesis examined the development, deployment, and evaluation of AuroraGPT, an AI-powered assistant for Northern Lights guidance. The study aimed to determine how such an assistant could be developed, identify the information required for effective guidance, assess how this data could be engineered into a system prompt for a foundation model, and evaluate whether the assistant could deliver near-expert-level guidance.
The knowledge base for this research included the historical development of Northern Lights guidance in the Rovaniemi area since the 19th century, the author's extensive experience with Northern Lights guidance, and key concepts and challenges related to Northern Lights visibility. A proof-of-concept was demonstrated using a local Python prototype. The primary development challenge was deploying a web application, which required full-stack development skills. This was addressed using the vibe coding method through the Lovable AI platform, enabling rapid deployment with a chat interface connected to the OpenAI API. However, this approach introduced challenges in debugging and codebase comprehension. The assistant's knowledge base was constructed around key concepts of Northern Lights guidance. The data engineering lifecycle framework supported data ingestion, storage, and transformation, allowing the assistant to automatically utilize high-quality data. Prompt engineering was the primary AI engineering technique.
AuroraGPT was evaluated using 14 test instances against a generic AI assistant (ChatGPT) and a traditional Aurora application (Aurora Compass). A baseline response was established for comparative analysis. AuroraGPT outperformed the other systems, delivering high-quality responses and user instructions. The results demonstrated AuroraGPT's capacity to provide near-expert-level Northern Lights guidance, with ongoing development planned.
The knowledge base for this research included the historical development of Northern Lights guidance in the Rovaniemi area since the 19th century, the author's extensive experience with Northern Lights guidance, and key concepts and challenges related to Northern Lights visibility. A proof-of-concept was demonstrated using a local Python prototype. The primary development challenge was deploying a web application, which required full-stack development skills. This was addressed using the vibe coding method through the Lovable AI platform, enabling rapid deployment with a chat interface connected to the OpenAI API. However, this approach introduced challenges in debugging and codebase comprehension. The assistant's knowledge base was constructed around key concepts of Northern Lights guidance. The data engineering lifecycle framework supported data ingestion, storage, and transformation, allowing the assistant to automatically utilize high-quality data. Prompt engineering was the primary AI engineering technique.
AuroraGPT was evaluated using 14 test instances against a generic AI assistant (ChatGPT) and a traditional Aurora application (Aurora Compass). A baseline response was established for comparative analysis. AuroraGPT outperformed the other systems, delivering high-quality responses and user instructions. The results demonstrated AuroraGPT's capacity to provide near-expert-level Northern Lights guidance, with ongoing development planned.