Skip to content

SeFuSe: semantic search over funding databases without external services

Many search tools for funding databases transmit the submitted project descriptions to commercial providers. For unpublished research ideas this is undesirable.

The Semantic Funding Search (SeFuSe) project addresses this constraint: the system runs entirely on local hardware, so that no query data leaves the operator's infrastructure.

What is SeFuSe?

SeFuSe identifies relevant funding programmes in the German Federal Funding Database (Förderdatenbank des Bundes) on the basis of a natural-language description of the project.

Instead of manual browsing, SeFuSe embeds the description with a local embedding model and ranks the programmes by the similarity of their embeddings to it, returning short summaries and direct links to the official entries.

Watch the demo

Watch the video

How it runs

SeFuSe is built as a modular, open-source stack:

  • FastAPI for the backend
  • Ollama for local embedding model inference
  • Qdrant as the vector database
  • Streamlit for the user interface

Everything is orchestrated with Docker Compose, so that deployment and updates require a single command and all data remains on the operator's own infrastructure.