Prisca-GraphRAG and Tawa Sparse Autoencoder: Lineage-Aware Retrieval with Governed Feature Interpretability

Key Points

• The central aim is to explore lineage-aware retrieval methods and integrate feature interpretability within frameworks using differential privacy.
• Developed Prisca-GraphRAG for lineage-aware knowledge retrieval
• Implemented Federated Prisca Privacy with differential privacy principles
• Created Tawa SAE for interpretive analysis of sparse autoencoders
• Established effective lineage-aware retrieval mechanisms
• Demonstrated improved feature interpretability
• Validated approaches for privacy in AI contexts