Zitha Sasindran

I am an AI engineer and researcher based in Auckland, New Zealand. I build AI systems that work on messy, real-world data and hold up once they’re deployed.

My work spans speech recognition, computer vision, federated and distributed learning, and large language models. I did my PhD at the Indian Institute of Science (IISc), Bangalore, on making machine learning work on resource-constrained devices, and led the Boeing-sponsored project that delivered an offline speech-to-text system for noisy aircraft cabins.

These days I focus on making AI trustworthy. I have created LLM-based evaluation metrics (SeMaScore and H-eval), stress-tested production LLMs and AI agents at HandShakeAI, and co-authored StepShield, on when to intervene on rogue AI agents. As an independent reasearcher, I have built LLM agents and RAG systems, with evaluation and guardrails built in from the start.

I also review for NeurIPS, ICASSP and INTERSPEECH, and enjoy mentoring engineers. I’m open to AI engineering and applied research roles in New Zealand, so feel free to reach out.

If you’ve got a fun project or want to chat, I am always open to new ideas-feel free to reach out!