Multi-agent LLM security framework. Cross-service attack detection and runtime monitoring, built to pressure-test the runtime threat models AI security platforms have to defend against.

Second place, Protegrity 2026 AI Pipeline Security Challenge. Personal data is tokenized on ingest and stays tokenized through embedding, the vector store and inference, so a successful prompt injection walks away with tokens instead of personal data. Plaintext re-identified 100 percent of records on real embeddings; AEGIS re-identified none.

Automated scanner testing AI endpoints against the full OWASP LLM Top 10. Semantic detection catches paraphrased attacks that bypass pattern matching. 35+ payloads, production-ready.

Multi-cloud misconfiguration scanner mapping findings to CIS Benchmarks with executive PDF reporting. Covers AWS, Azure, and GCP in a single read-only scan.