Hi, I'm Kourosh.

I expanded an undergraduate software-engineering project into my B.Sc. thesis, developing a Git-based workflow for incremental code-quality review using large language models to support human evaluators. The workflow combines project requirements, evaluation criteria, domain documentation, and code changes with evidence validation and traceable reporting. I evaluated it on real Java and Go projects and compared different ways of providing project-specific context.

I study building-level damage assessment from pre- and post-earthquake remote-sensing imagery to support faster identification and prioritization of affected structures after major earthquakes. The goal is to support faster field inspection and resource allocation during disaster response. I designed and evaluated an approach that first identifies affected buildings and then estimates damage severity. Across four earthquake events and 373 test tiles, it reached affected-building precision and recall of 61.07% and 33.81%, compared with 53.66% and 25.98% for the reference baseline on common support.

I design LLM-simulated Persian student personas with bounded knowledge, stable misconceptions, confidence, engagement, and help-seeking behavior to test AI tutoring systems before enough real student dialogue is available. I also built a 5-10-turn LMS evaluation workflow that records conversations and session metadata to assess persona consistency, misconception stability, and tutor responses across student profiles. The resulting dialogue data supports controlled evaluation and future personalization.