Join us for our next network seminar! Read more about it
No items found.
.jpg)
Power grid companies face a high strain on customer service, especially in the event of incidents such as power outages or invoice sending. Customers expect quick and precise answers, and more want to be more self-reliant and drop phone calls. How can power grid companies improve the customer journey by reducing latency, increasing availability and providing more precise information, while maintaining cost-effective operations? What kinds of technological solutions and processes can be implemented to meet customer needs without increasing costs or resource use?
Through workshops with the customer, we explored how KI can help users as a first step in solving specific challenges. We quickly developed a proof-of-concept based on a RAG solution with LlamaIndex, trained on real website data and documents. The results showed great potential and formed the basis for further development.
The next step was to make the solution production-ready with a focus on scalability, security, and a clear design strategy. We implemented data security measures, spam filtering and evaluated the model with tools such as RAGAS to optimize configuration. Users were also given the opportunity to upload their own documents, such as invoices, to the KI Assistant.
The design built on Gture's own KI framework, and was iteratively improved based on feedback. To monitor usage, we used LangFuse as a basis and developed an extended version with, among other things, automatic categorization and customer score logging.
Eventually, all components—RAG, analysis tools, and design—were linked together in a scalable system in AWS. Launched in November 2024, Netti is now being rolled out to more power grid companies, with the goal of reaching over 20 companies nationally. Continuous use and feedback provide the basis for further improvements and scaling.
An AI assistant as the first line support for online rental customers that efficiently handles common outage inquiries, invoices and general queries. The KI assistant is adapted to specific industry knowledge and data from online companies in Norway. It is trained on relevant documents and information relating to various online rental companies, and can thus provide precise and relevant answers to customers in different locations. This solution increases availability, reduces wait times and empowers customers to get help without having to contact customer service directly, while maintaining cost-effective operations.