German Federal Ministry of Research, Technology, and Space (BMFTR)

Challenge

The German Federal Ministry of Research, Technology, and Space (BMFTR) manages several million euros in annual project funding. BMFTR staff responsible for project funding must adhere to complex regulations contained in a 200-page Project Funding Manual. While available on the BMFTR intranet, finding information was time-consuming and difficult. BMFTR sought to use AI to simplify internal workflows and boost efficiency by providing quicker, more intuitive access to the manual's contents.

Solution

In just six months, a collaborative team including BMFTR, deepset, our partners HalloWelt and GovTech Campus, and the Fraunhofer Institute, developed a chatbot that can answer questions about project funding rules in natural language. The chatbot:

• Meets high data protection and usability requirements
• Complies with all aspects of the C5 cloud security criteria
• Takes less than three seconds to answer each question

Technical Architecture

The team implemented a RAG (Retrieval Augmented Generation) architecture, a technique that grounds AI responses in factual source material. This approach:

• Ensures responses are based exclusively on approved documentation
• Provides source citations for every answer
• Eliminates the risk of hallucinations or fabricated information

User-Centered Design

The team worked in close collaboration with 20 end users, ensuring the chatbot and user interface addressed real-world use cases effectively. 

"By involving the users from the beginning and focusing on the UX, we were able to create a successful product that has been well-received," says project manager Josefine Theden-Schow.

Impact

The chatbot has become an essential daily tool for BMFTR staff, with measurable benefits:

• 1000 queries conducted weekly without requiring additional staff resources
• Reduction in issue resolution time
• Enhanced transparency through verifiable source citations and references

“Thanks to the collaboration with deepset, we were able to develop a finished product in just six months that meets our agency's stringent requirements for information security and accuracy. We have learned a lot about developing products with artificial intelligence and have positioned ourselves for the future."- Josefine Theden-Schow, Project Manager, BMFTR

This project serves as a model for government bodies that want to share complex knowledge with staff or the public, reduce processing times, and ensure consistent application of regulations.

Why deepset?

deepset is an outcome-focused AI leader that helps enterprises, public sector, and defense build secure AI applications—fast and fully customized to their highest-value use cases. Since 2018, we have helped customers such as Airbus, OakNorth Bank, and The Economist to develop production-ready AI products. The deepset AI Platform uses our popular open-source Haystack framework for AI orchestration and combines our many years of experience in building AI products tailored to different use cases and industries.