SeFuSe: an open-source semantic search for funding programmes

Identifying a suitable funding programme for a research idea or product concept is a time-consuming task. SeFuSe is an open-source tool, operable on a local machine or an institutional server, that supports this search.

Semantic Funding Search (SeFuSe) locates relevant programmes in the German Federal Funding Database from a natural-language description of the project. Rather than requiring the user to guess suitable keywords, it compares the embedding of the description with the embeddings of the programme descriptions and returns the nearest matches.

How it works

1. The project description is entered in the web interface.
2. A local embedding model converts it into a vector, which is matched against a vector database of funding programmes.
3. The matching programmes are returned with summaries and direct links to the official entries.

Technical stack

• FastAPI: backend and data processing
• Ollama: local embedding model inference
• Qdrant: vector database for semantic search
• Streamlit: web interface
• Docker Compose: orchestration of all services

All of these components run on the operator's own hardware, so no query data leaves that machine.

Rationale

• Project descriptions are not transmitted to commercial providers.
• The system is open source and modular, and can be deployed without external dependencies.
• The index of funding programmes is refreshed on a regular schedule.

Short demo