Development of a Vector Database for Long-Term Memory Storage in Foundation AI Models

07/2024 – 12/2027 | SW Computing Core Technology R&D Program, IITP

• Development of ultra-high-performance vector database (Vector DB) and indexing technologies for long-term memory storage in generative AI
• Implementation of techniques for accuracy-guaranteed retrieval, including automatic parameter tuning, out-of-distribution adaptive graph indexing, and multimodal query handling
• Integrated design of both in-memory and disk-based vector DB, with Korea’s first demonstration of multimodal LLM–NPU integration
• Open-sourcing of core vector DB technologies and promotion of Apache project registration to enable global adoption and industrialization