City Council of Valencia — Information and Communication Technologies Service (SerTIC).
Public Administration
Conversational AI Assistant for internal support and IT incident management
Large organizations manage a high volume of internal tech queries and incidents daily, many of which relate to recurring issues such as passwords, VPN access, office tools, or corporate procedures.
In an administration like the City Council of Valencia, this activity represents a significant workload for first-level support teams, who must address similar queries while managing incidents that require more specialized intervention.
In this context, the Information and Communication Technologies Service (SerTIC) proposed developing an AI-driven conversational assistant capable of enabling employee self-service, leveraging the organization’s accumulated knowledge, and automating the resolution of frequently asked questions without losing control over those requiring human intervention.
Challenge
The main challenge was to provide a fast, reliable response to employees’ IT queries while reducing the operational burden on Tier 1 support.
The solution needed to understand questions phrased in natural language and locate the relevant information across manuals, procedures, internal regulations, and ticket histories—avoiding generating answers not backed by available knowledge.
Additionally, a mechanism was required to detect when a query could not be resolved automatically and escalate it to the support team, retaining the necessary context to continue its management. The assistant had to integrate into internal support channels and accommodate queries via text, images, or voice.
Solution
Foqum developed a conversational AI assistant that allows employees to ask technology questions in natural language and receive answers sourced from the City Council’s internal knowledge base.
The system combines semantic search and Retrieval-Augmented Generation (RAG) over manuals, procedures, regulations, and ticket histories. The model generates answers strictly from this content, incorporating traceability, evaluation, and confidence-control mechanisms to enhance response reliability.
When the system lacks sufficient information to resolve a query, it recognizes its limits and automatically generates an incident in Redmine, transferring the context and necessary data for the support team to take over. The solution also supports attaching images and making voice queries by transcribing the request and providing audio output for the response.
