Interview with Sukanya Mandal: Synthesizing Multi-Modal Knowledge Graphs for Smart City Intelligence

Cognitive digital twins (CDTs) are emerging as the next evolution of urban management, offering AI‑powered virtual replicas that simulate the interplay of physical infrastructure, human behavior, and environmental factors. Traditional digital twins often rely on siloed, structured datasets, limiting their ability to capture the nuanced, cross‑domain relationships that define modern cities. Knowledge graphs provide the semantic backbone needed for explainable reasoning, linking traffic sensors, health records, energy usage, and social media streams into a coherent, queryable model. By embedding these diverse data sources, CDTs can forecast congestion impacts on emergency services or evaluate energy demand under varying weather conditions.

The LLMasMMKG framework tackles the data integration challenge by leveraging large language models at each stage of graph construction. First, heterogeneous inputs are harmonized using Sentence‑BERT embeddings, creating a shared semantic space for text and time‑series signals. Fine‑tuned BERT extracts entities, while GPT‑4 generates relational triples, all of which are stored in an RDF graph aligned with a hierarchical ontology. Crucially, the pipeline synthesizes realistic data—textual descriptions, sensor readings, and simulated patient notes—using GPT‑4‑turbo, sidestepping privacy concerns and filling gaps where real data are scarce. This synthetic‑first approach not only reduces manual curation costs but also offers a reproducible testbed for researchers and city officials.

Beyond the proof‑of‑concept, the work signals a broader shift toward privacy‑by‑synthesis and neurosymbolic AI in urban planning. Upcoming enhancements such as cross‑modal entity linking, incorporation of image and video streams, and graph‑neural network reasoning will deepen the twin's predictive capabilities. For municipalities, the ability to rapidly generate and update multi‑modal knowledge graphs means faster scenario analysis, more transparent policy decisions, and a scalable path toward sustainable, data‑driven cities. As smart‑city initiatives mature, frameworks like LLMasMMKG could become the standard infrastructure for building resilient, intelligent urban ecosystems.