At Ciena, I fine-tuned ModernBERT models for semantic Block/Mask/Allow classification, achieving 97–99% accuracy with the base model across security tests. I also engineered multi-agent orchestration and LLM guardrails to block prompt injection and harmful payloads.
I built a production Qdrant service with a Rust wrapper and high-availability features for retrieval across 7 microservices. I also standardized Helm deployments for 128 microservices across BP2 and Kubernetes, and developed an agent evaluation system using LangGraph, WebSockets, and the A2A protocol.

