Portfolio

Semantic Knowledge Graphs

Traditional knowledge graphs aim to extract and structure information from general text. Our semantic knowledge graph generation approach leverages not only embedding models used in retreival augmented generated but also link prediction methods to create dense graphs.

Applications include:

  • Internal knowledge management and discovery

  • Information discovery such as legal case files

  • Project management and organisation overviews

SemaDB

SemaDB is a hybrid multi-vector multi-index vector search engine designed to be easy-to-use. It was initially concieved as a component for knowledge management projects. It is now an open-source project with cloud hosting.

Applications include:

  • Retrieval augmented generation

  • Semantic similarity search

  • Hybrid text and vector search

From pixels to logical rules

This work presents a novel method to obtain logical rules directly from end-to-end differentiable neural networks. The overall method can learn logic programs on top of neural perception such as convolutional neural networks.

Applications include:

  • Post-training inference intervention

  • Logical grounding of neural networks