Safi Shamsi

Shamsi studied data science at the University of Birmingham. His published thesis repository combines a Neo4j graph of papers and authors with text embeddings and a retrieval-augmented language-model pipeline. The graph supplies structured relationships while text retrieval finds semantically related material; retrieved documents ground the resulting response. The repository identifies the work as a 2025 master’s thesis project. Its reported performance improvements are project measurements, not a guarantee that another dataset or application will produce the same results.[4]

Graphify represents a codebase as connected entities and relationships rather than a collection of text fragments to search repeatedly. An assistant can follow those relationships to find relevant functions and dependencies, then trace an answer back to source material. Graphify Labs presents that graph as a shared memory layer for engineers and coding agents, with an enterprise layer for review and supported verification workflows. Shamsi’s founder biography connects this work to his master’s research on knowledge-graph-powered retrieval.[3][2]

The Memory Layer develops the argument behind Graphify for people choosing or building AI systems. It follows graph theory, ontologies and retrieval through the problem of an assistant that loses useful context between sessions, then connects those ideas to a buildable memory pipeline. The publisher offers both conceptual explanations and code. It is a useful route into the design choices behind the software before attempting to evaluate a production deployment.[5]