Vippsi Gets Smarter: Expanding File Support

Hey AI enthusiasts! 👋

I’ve been heads down working on Vippsi, our internal AI-powered conversation tool, and wanted to share what I’ve been up to. Imagine having a natural conversation with your entire organization’s knowledge base - asking questions about policies, documentation, or reports, and getting instant, contextual responses. Instead of scanning through hundreds of pages, you can simply ask questions in plain language and get precise answers drawn directly from your documents.

As a financial services company, Vipps MobilePay faces unique challenges with AI adoption. While solutions like OpenAI, Anthropic, or Google offer powerful capabilities, they aren’t always suitable for our needs. Financial institutions require stringent compliance measures and data privacy guarantees, while keeping costs manageable at scale. That’s why we’re building Vippsi as a custom solution that puts privacy and compliance first.

Work in Progress

We’re expanding Vippsi to support multiple file formats:

  • PDF documents
  • Word documents (DOCX)
  • PowerPoint presentations (PPTX)
  • Plain text files (TXT)
  • More formats coming soon

Technical Architecture

Let me explain how we’re handling your documents behind the scenes. When you upload a file, it’s processed in our backend but - and this is crucial - we never store anything. Think of it like having a conversation with someone who can read your document, answer your questions, but immediately forgets everything once you’re done.

Your files and conversations stay right in your browser using IndexDB storage. When you chat about a document, we convert its content to base64 format (a way to represent binary data as text) and stream it back to your browser. This base64 version gets stored locally and is used in your future conversations, making subsequent interactions quick and efficient while keeping everything on your device.

Here’s where it gets clever: Each time you ask a new question, your browser sends the base64 encoded content, which is much smaller than the original document. No need for storage on our end, no conversation history to worry about - everything stays private and under your control.

Current Status & Next Steps

The backend architecture is ready, while frontend work continues:

  • Optimizing file upload experience
  • Implementing streaming UI feedback
  • Fine-tuning document conversations
  • Adding support for more formats
  • Enhancing conversation quality

Stay tuned for more updates as we continue to enhance Vippsi’s capabilities!

~ Sven