The knowledge has always been there. Now it can be found, too.
VAGO has developed a customized RAG architecture for FRICKE Abfülltechnik that enables access to corporate wikis and email communication via natural language—quickly, precisely, and without detours.
Challenge: Hidden Knowledge
In a manufacturing company like FRICKE Abfülltechnik, up-to-date information isn’t just a nice-to-have. It’s essential for smooth operations. Technical inquiries, process documentation, relevant email threads—all of this existed within the company, but it was nearly impossible to find reliably through manual keyword searches.
This led to a familiar pattern: Relevant information was either not found or took a considerable amount of time to piece together. Employees asked colleagues instead of looking it up in the system. And the results they did find were often imprecise or outdated.
Solution: Semantic Search Instead of Guessing Keywords
VAGO developed a powerful RAG architecture specifically tailored to FRICKE’s information needs and system landscape. The solution is directly integrated with the existing wiki and email systems. All relevant content was stored in a database for semantic searches and enriched with process-relevant metadata.
Employees now formulate their queries in natural language. Instead of searching for individual keywords, they describe what information they need and in what context they need it. The system understands the query, filters out the relevant portion of the data based on metadata, and delivers a precise, transparent answer.
Technical Highlights: Efficiency as an Architectural Decision
A key design feature of the solution is the two-stage search. Before a semantic search is performed, metadata filtering narrows down the dataset to be searched to the truly relevant portion. This increases the precision of the results and keeps computational overhead low. The AI models used are as small as possible and as large as necessary. At VAGO, cost efficiency and process performance go hand in hand.
- Two-stage search with metadata filtering
- Seamless integration with existing wiki and email systems
- Resource-efficient model selection based on process requirements
Result: Faster answers. Better decisions.
Employees at FRICKE can now find relevant information from wikis and email threads in seconds. The tedious keyword search is a thing of the past. This not only saves time but also reduces errors caused by missing or misinterpreted information. The company’s knowledge is finally working for the company.