Hundreds of tickets. No manual effort.

Challenge: A government agency overwhelmed by daily ticket volume

The ZSVR receives numerous inquiries every day—legal questions, technical support, and general concerns. Until now, these tickets were reviewed manually, categorized by topic, and answered by staff members via email. The system worked, but it was slow, labor-intensive, and offered no room for scaling.
The real problem wasn’t just the volume. It was the time that qualified employees had to spend on repetitive classification and drafting tasks instead of focusing on more challenging work.

Solution: An agent-based pipeline for the entire process

VAGO, in collaboration with ZSVR, developed a fully automated ticket-processing pipeline that handles inquiries completely autonomously, from classification to context-sensitive responses. The system consists of two central AI components: a classifier model trained on real ZSVR ticket data that assigns incoming inquiries to the correct category with over 90 percent accuracy, and a finely tuned LLM that responds to the classified tickets. To do this, an agent accesses the ticket content, retrieves relevant information from internal databases via RAG, and formulates a tailored response. If the ZSVR requests human review before the response is sent, the system makes the response available for review. If no review is required, the pipeline sends the response directly.

Technical Highlights: Seamlessly Operated, Resource-Efficient Implementation

The entire solution was developed, trained, and operated on ZSVR’s own servers. Sensitive government data never left the internal network at any time. The system runs on off-the-shelf GPUs and is powered by models ranging in size from one to eight billion parameters. Resource efficiency was not a compromise but a design principle.

• Developed and operated entirely on-premises
• Classifier with over 90% classification accuracy
• Runs on off-the-shelf GPUs without specialized infrastructure

Result: Less manual effort. More capacity for what matters most.

Thanks to the automated pipeline, the ZSVR saves a significant amount of manpower in daily ticket processing. Case workers are relieved of repetitive classification and drafting tasks and can focus on processes that require genuine professional judgment. The system is on-premises, GDPR-compliant, and fully integrated into the existing Jira infrastructure.