At Ejada Systems, I developed a function-calling evaluation pipeline for LLM agents that improved agent accuracy by 30%. It validates tool selection, parameter extraction, and execution reliability.
I architected an agentic RAG system using LangGraph and two-stage retrieval, and designed a three-tier guardrails framework. I also fixed a client data-leakage bug in the proposal service.
I build and maintain production AI microservices for a wealth-intelligence platform serving a Saudi capital-markets client. The work includes churn prediction, next-best-action recommendations, vector search, SHAP explanations, and LLM-driven proposal generation.
Earlier, as a Computer Vision Intern at Technocolabs Softwares, I built ResNet-50 transfer-learning classifiers for inventory reconciliation. My projects include multimodal fashion search and EEG-based motor-intent prediction for prosthetic arm control.
