<- Back to projects

Protegrity ยท Summer 2026

Privacy-Preserving Knowledge Graph Search

A semantic search system for sensitive knowledge graphs, designed so useful retrieval and LLM reasoning can happen without exposing plaintext data.

What I built

I designed and implemented an encrypted retrieval pipeline using CKKS homomorphic encryption and CPE pseudonymisation. The system supports vector search over protected data and passes relevant results to an LLM without revealing the source records to an untrusted environment.

Impact

  • Achieved a 25x speedup over the initial encrypted retrieval flow.
  • Reached 100% retrieval accuracy on the evaluation set.
  • Completed a Qiskit and quantum computing bootcamp alongside the project.