Incremental Reasoning over Evolving Knowledge Graphs
KAIROS is a research project at USI studying how AI agents can reason over distributed, continuously changing knowledge sources. The project explores incremental graph processing, dependable storage, and interfaces for LLM-based reasoning.
Research Questions
Incremental Graph Processing
Investigating how evolving knowledge graphs can be queried without recomputing complete traversals from scratch.
Dependable Distributed Systems
Exploring replicated, fault-tolerant architectures with strong consistency guarantees for graph-based state.
Interfaces for Agentic AI
Studying how LLM-based agents can access structured, changing data during multi-step reasoning workflows.
Core Research Directions
Neighborhood Prefetching
Studying whether graph neighborhoods can be prefetched to reduce latency during iterative reasoning.
Predictive Access
Exploring how an agent’s interaction history can inform the next likely portions of the graph to access.
Traversal Optimization
Investigating techniques to reuse intermediate state and avoid unnecessary repeated graph traversals.
Who KAIROS connects with
Distributed Systems Researchers
Researchers interested in replication, partitioning, consistency, fault tolerance, and state management.
AI & Knowledge Graph Researchers
Researchers studying LLM grounding, GraphRAG, agentic AI, and reasoning over evolving structured knowledge.
Potential Collaborators
Students, labs, and industrial partners interested in experimental evaluation and real-world data scenarios.
Discuss Research Collaboration
KAIROS is conducted by PhD students, researchers, and faculty at the Distributed Systems Lab at USI. Contact us for proposing research collaborations or to receive research updates, publications, prototypes, and calls for collaboration.