About me

I’m Neeraja Kirtane, my research is focused on building trustworthy and interpretable language models. My work lies at the intersection of AI safety, robustness, and interpretability, with the goal of making large language models (LLMs) more reliable and transparent.

I’m currently a Research Engineer at MathGPT.ai, where I’m developing education-centric reasoning benchmarks to evaluate reasoning robustness in state-of-the-art models. My work explores how simple linguistic or contextual changes can destabilize model reasoning — and how fine-tuning small language models (SLMs) can improve their consistency and usefulness in AI tutoring and educational applications.

Alongside this, I am working under the supervision of Prof. Kuan-Hao Huang at Texas A&M University to probe reasoning and multilingual generalization in LLMs. Using interpretability tools such as neuron activation analysis and feature attribution, we are studying how models encode reasoning processes across languages and how these representations transfer between linguistic systems.

Before this, I completed my M.S. in Computer Science at the University of Illinois Urbana–Champaign (UIUC), where I was advised by Prof. Hao Peng and Prof. Dilek Hakkani-Tür. At UIUC, I worked on projects aimed at improving trust and accountability in LLMs:

• FactCheckmate – a framework for preemptively detecting and mitigating hallucinations in LLMs by analyzing their hidden-state dynamics and identifying early indicators of factual inconsistency.

• Jailbreaking LLMs – a study of scientific-sounding adversarial prompts that can elicit biased or toxic model responses, revealing deeper vulnerabilities in instruction-following and safety alignment.

Previously, I worked as a Research Assistant at IIT Madras under Prof. Balaraman Ravindran and Dr. Ashish Tendulkar

I’m currently seeking PhD opportunities or full-time research roles in AI safety, robustness, and interpretability.

If you’re interested in my work or would like to collaborate, feel free to reach out via Email or LinkedIn. You can also view my CV for more details.