LuminaSwiss

LuminaSwiss: secure, private LMS that lets organizations search and learn from their own documents safely with Swiss hosted LLMs

Comment

LuminaSwiss — Project Story

We built LuminaSwiss as part of our journey at Digitalizers Geneva, a hands-on training environment designed to strengthen digital skills, agility, practical tool use, collaboration, and employability. That context mattered a lot: we were not just trying to deliver a demo, we were trying to prove that we could work like a real product team and turn learning into a credible solution.

What inspired us

Our inspiration came from a simple tension we kept seeing: organizations want to use AI to make learning and knowledge access faster, but they do not want to lose control of their internal documents or create anxiety for privacy and compliance stakeholders. We wanted to build something that answered both sides of that equation:

[ \text{Useful AI} + \text{Controlled Sources} + \text{Privacy by Design} = \text{Trustworthy Adoption} ]

What we built

LuminaSwiss is a controlled and secure AI-powered learning management system that allows organizations to query their own documents through a privacy-first RAG workflow.

The product backlog helped us shape that vision into concrete functionality: secure authentication and session control, document upload, PII detection and tokenization, text extraction, embeddings and semantic retrieval, LLM integration, HTTPS/TLS through Caddy, and sanitization at the request and response levels. The result is a tool designed for organizations that want AI-assisted knowledge management while keeping their Data Protection Officer reassured.

How we built it

We built LuminaSwiss like a real agile team. We worked from a backlog, ran ceremonies, tracked progress on a Kanban board, and gradually reduced the scope toward an MVP that could still make a strong impression. On the technical side, we designed the system around a browser-based interface, secure API access, controlled document ingestion, RAG-based retrieval, and protected communication between components. We initially explored the idea of a local LLM on the backend, then adapted the architecture and moved to an API-based approach when that became the more realistic and robust path for the project.

That pivot taught us an important lesson: architecture and tech-stack choices are not secondary decisions. They shape everything that follows.

Challenges we faced

This project was ambitious for our context. The team started with very basic agile knowledge, we also started late, which increased the pressure to make good decisions quickly.

Our biggest challenge was balancing security, scope, and showcase value. We did not want to build just another AI demo. We wanted to demonstrate a convincing security concept while still delivering a working MVP. That forced us to make trade-offs constantly: what had to be in the first version, what could be simplified, and what had to remain visible to prove the concept. The move from a local-LLM idea to an API-based approach was part of that reality: we had to stay flexible and adapt to what would best serve the product and the time available.

What we learned

We learned that:

  • choosing the right architecture early saves time later,
  • a backlog becomes powerful when it is actually used to drive decisions,
  • agile ceremonies matter more when a team is still learning how to work together,
  • and collaboration can accelerate learning dramatically.

We also learned a lot from one another, and from AI itself, while implementing the project. Technical topics that might have stayed abstract in theory became concrete because we had to make them work together in one coherent system.

Why this project matters to us

LuminaSwiss is more than a prototype for us. It is proof that, as trainees, we can move from learning concepts to delivering a structured, relevant, and professional solution. It reflects the spirit of Digitalizers well: practical learning, agility, human-centered digital transformation, and a strong link between skills development and real market needs.

What's next for LuminaSwiss

Our next goal is to turn LuminaSwiss from a working MVP into a credible pilot for organizations. We want to strengthen the end-to-end secure RAG flow, improve transparency and admin control, and test the solution on a concrete use case such as internal training, onboarding, or compliance knowledge management.

We are proud of the result not only because of the tool itself, but because of the way we built it: as a team, with discipline, adaptability, and a genuine desire to create something useful.

Built With

Share this project:

Updates