Lasse Hyldahl Jensen

In early 2024, two colleagues and I launched Lunar’s AI colony to explore and implement generative AI across the company’s products and business processes. The colony’s mission was to gain hands-on experience with generative AI and demonstrate its potential to drive rapid value for the business.

In our first month, we focused on creating internal tools, starting with an internal chatbot, followed by a customer support chatbot that could access Lunar’s knowledge database. These tools proved highly effective, allowing us to focus on building a customer-facing AI assistant.

In April 2024, we launched the AI assistant. Following its release, we gathered user data and feedback to refine the assistant’s functionality. Once we had established organisational trust in customer-facing AI solutions, we expanded to build a more comprehensive AI support feature.

The AI support system, based on the foundation of the AI assistant, is specifically designed to handle customer inquiries. Integrated with Lunar’s customer support platform, Intercom, via Intercom’s REST API, this solution allows customers to receive instant responses rather than waiting up to three days for replies. Today, the AI support system independently resolves 60% of customer inquiries. For the remaining 40%, it gathers relevant information—such as screenshots of technical issues—before escalating to a human representative.

Together, the AI support system and AI assistant manage over 40,000 customer interactions monthly. All AI solutions are built with Lunar’s tech stack structured as Go-based microservices with Swagger and GraphQL APIs. The RAG implementation uses a PostgreSQL database with the pgvector extension to enhance search and retrieval capabilities.

At Lunar, I develop microservices in Go, focusing on services that expose APIs to Lunar’s app and internal customer management system via Swagger and GraphQL.

During my internship, I primarily worked on implementing an anti-corruption layer (ACL) within a microservice that retrieves data from one of Lunar’s partners. Previously, API calls were directly forwarded to this partner, which introduced latency and limited control over response times. The ACL mitigates these issues and enables event sourcing across Lunar’s microservices, allowing greater flexibility and control over partner data.

As part of Lunar’s DevOps strategy, updates to this microservice were continuously released to production, sometimes multiple times per day. Today, these APIs are called each time one of Lunar’s 950,000 users opens the app.

In my full-time role, I have taken on the responsibility of updating Lunar’s card management system. Originally developed when Lunar received its banking license in 2019, this system carried technical debt due to time constraints for the 2020 launch, as the platform had not yet matured. The update primarily focused on resolving architectural issues that limited scalability for future product launches. Additionally, I implemented an integration with Lunar’s card issuer, Nets, to enable synchronization between the two systems and automate handling of any discrepancies.

For my Bachelor’s project, I collaborated with IBM Denmark and Smukfest to develop the backend for Smukfest’s new mobile app. The backend was designed with a microservice architecture to ensure scalability and maintainability, and it integrates various Content Management Systems (CMS) while enabling push notifications to all devices via Expo.

The app successfully launched for Smukfest 2023, supporting over 50,000 attendees, and was designed with resilience, ensuring continuous functionality despite backend errors.

The Playback Controller, an Embedded Linux device for intelligent architectural lighting, is configured through a React-based app over Wi-Fi, Ethernet, or mobile network. Users can set up static and dynamic lighting scenarios, triggered by time, sunrise/sunset, SMS, or Moto Muto’s website, with IBM Node-RED logic embedded in the controller.

When connected to the internet, the Playback Controller uploads logs to Moto Muto’s Elastic Stack, where they are analyzed and displayed in a Kibana dashboard. The device is also used in Moto Muto’s “Sun Shelf” installations. I developed all software for the Playback Controller and its supporting infrastructure.

The System Controller is Moto Muto’s earlier lighting control system, used in settings like hospitals, nursing homes, and psychiatric facilities where circadian lighting is essential. It communicates with Moto Muto tablets via WebSockets and with Moto Muto Zone Controllers via DMX/RDM protocols.

I contributed to the development of RDM communication between the System Controller and Zone Controllers, allowing the system to display events from Zone Controllers on Moto Muto tablets. I was also responsible for ongoing software maintenance.