Context
Before it became possible to casually chat with PDFs inside ChatGPT, there was a gap for people who wanted a fast, local, native experience to deeply interact with research papers. ViewAI was built to fill that need.
What It Was
ViewAI was a native macOS app designed by Kryštof Mitka. It let users open any scientific paper (PDF), view it cleanly in a two-panel layout, and instantly ask questions about the content using a GPT-4-powered retrieval-augmented generation (RAG) stack.
It was built during a time when no easy, off-the-shelf tools existed to achieve this with speed, quality, and local control. Everything from parsing to vector search had to be custom built.
Features
- Local parsing of PDFs using Nougat for structure extraction
- LaTeX math rendering support
- Embedded vector search on-device using scientific text chunks
- GPT-4 used for answering questions via RAG architecture
- Clean native UI/UX in Swift, focused on speed and simplicity
Tech Stack
- Swift (native macOS)
- Nougat (for PDF to structured data)
- LaTeX rendering engine
- GPT-4 API
- Custom vector embedding store
Outcome
ViewAI served as an early glimpse into what user-friendly, AI-native interfaces could look like for academic work. It helped validate the demand for structured, smart interactions with scientific content—long before the mainstream tooling caught up.
This was a small but sharp experiment that Kryštof led and executed, showing what was possible with the right combination of AI infrastructure and product intuition.