Basmalla Metwally

I build hands-on AI projects, including multi-agent LLM systems with LangGraph and RAG, while studying Artificial Intelligence at Horus University.

In my AI DataAnalystAgent project, I designed a LangGraph investigation pipeline where a CriticAgent adversarially reviews conclusions against an evidence-weighted confidence floor. I added offline fallback mode, LLM caching, SQL-injection guardrails, per-user data isolation, and 42 offline CI tests.

I also built a Business Digital Twin and Decision Intelligence Platform that selects among four regression models and an ensemble, outperforming a naive MAPE baseline by 40%. Its causal elasticity estimates feed a 5,000-run Monte Carlo risk simulation, supported by placebo and random-common-cause refutation tests.

My work spans accessible full-stack products as well: AccessAI Companion uses React, TypeScript, and Express to support scene description, reading, live captions, and translation for blind, low-vision, Deaf, and hard-of-hearing users. I'm open to internship opportunities in AI, machine learning, and related fields.