A Privacy-Preserving Personal Knowledge Graph Using Local LLMs and Graph Databases

We highlighted the benefits of building a Personal Knowledge Graph based on our proposed concept of Privacy First Personal Knowledge Graph compared to traditional cloud-based data storage. We did that by carrying out a comparative analysis of Eidomuse, used for Contextual Memory Retrieval. In this instance, Eidomuse utilizes Graph-RAG technique for efficient linkage of data points that are disconnected, thereby allowing for natural language insights. Through the utilization of Edge Computing on consumer devices, we were able to create a privacy-preserving approach. Compared to what is in literature regarding AI adoption through clouds, we have shown that our model is superior when it comes to data sovereignty, trust, and contextualization. This research has opened up new avenues for actual implementation of AI within the personal space, leading to self-awareness, while still keeping privacy. The system might be able to be enhanced in several ways: ·          Domain Expansion: With the usage of domain ontologies, the scope of archiving can be further extended to include domains such as "Medical Health Records" or "Financial History." ·          Proactive Agents: Designing autonomous agents for scheduling a meeting or writing emails, depending on the graph context. ·          Biometric Integration: Adding smartwatches so as to relate physical manifestations of stress with work. Meetings or drafting emails using the context provided in the chart (example: "Draft a reply using my last project notes"). Other streams of data, such as Biometric Integration through smartwatch, can be introduced to connect physiological measures of stress with the pattern of activity in the digital realm (e.g., "Heart rate high during tight deadlines"). Lastly, the robustness of this system can be enhanced by introducing Local Re-ranking (LoRA), where the LLM can learn the personal lexicon and tone of voice of a user without external data. In summary, with advancements in technology, it is easy to incorporate new developments in Generative AI and Decentralized Web.  

[1]      Y. Cai, G. Zhang, Y. P. Pei, W. Bian, and W. Zheng, "SimGRAG: Leveraging Similar Subgraphs for Knowledge Graphs Driven Retrieval-Augmented Generation," in Proc. 63rd Annual Meeting of the Association for Computational Linguistics (ACL), 2025. [2]      M. Li, S. Miao, and P. Li, "Simple Is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation," arXiv preprint arXiv:2410.20724, 2024. [3]      L. Rossetto et al., "LifeGraph: A Knowledge Graph for Lifelogs," in Proc. Knowledge Graph Conference (KGC), 2020. [4]      L. Rossetto et al., "LifeGraph 4 – Lifelog Retrieval using Multimodal Knowledge Graphs and Vision-Language Models," in Proc. 7th Annual ACM Lifelog Search Challenge (LSC ’24), pp. 1-5, 2024. [5]      V. Kamalesakumar, V. Dhanush, S. Sathish, and M. Sathiyslakshmi, "AI-Powered Personal Knowledge and Content Management with Big-Brain," International Research Journal on Advanced Engineering Hub (IRJAEH), vol. 3, no. 4, pp. 1328-1335, 2025. [6]      Y. Guo et al., "FEEL: A Federated Edge Learning System for Efficient and Privacy-Preserving Mobile Healthcare," in Proc. Int. Conf. Parallel Process., 2020. [7]      N. Alasbali et al., "Privacy-Enhanced Skin Disease Classification: Integrating Federated Learning in an IoT-Enabled Edge Computing," Frontiers in Computer Science, vol. 7, pp. 1-12, 2025. [8]      C. Thiraviya et al., "A.I. Based Virtual Personal Assistant," International Journal of Advanced Research in Science, Communication and Technology (IJARSCT), vol. 5, no. 1, pp. 502-510, 2025. [9]      Edge, D., Trinh, H., Cheng, N., Bradley, J., Chao, A., Mody, A., Truitt, S., & Larson, J. (2024). From Local to Global: A Graph RAG Approach to Query-Focused Summarization. arXiv:2404.16130. Microsoft Research. [10] Han, H., et al. (2024). Retrieval-Augmented Generation with Graphs (GraphRAG). arXiv:2501.00309.