Aibak Aljadayah

At RPM International, I architected a production RAG pipeline using Azure OpenAI, GPT-4o, LangGraph, and Azure Cosmos DB that reached 89% ticket-classification accuracy across hundreds of daily production tickets. I replaced manual dispatch with an LLM workflow that reduced handling time from about 60 minutes to about four seconds.

I'm a graduate research assistant at Miami University, where I build LLM robustness testing systems, analyze transformer attention and neuron activations, and develop defenses that reduced attack success rates to 0% across five open-source models. My thesis work focuses on AI safety and transformer-based defense mechanisms.

I've also built multimodal content-safety, semantic-segmentation, and speech-emotion-recognition systems. My AvantiGuard pipeline outperformed evaluated baselines across four adversarial attack benchmarks and was containerized for real-time moderation through FastAPI and Azure Kubernetes Service.

I bring hands-on experience taking AI systems from model evaluation and retrieval design through observability, validation, APIs, and cloud deployment.