Hi! I’m Saba Ahmadi, an AI researcher working on post-training, reasoning, and parameter-efficient adaptation for language models.

In October 2025, I joined Noah's Ark Lab, where I work closely with Prasanna Parthasarathi and Yufei Cui. My current work focuses on post-training, reasoning, and parameter-efficient adaptation for both autoregressive and diffusion language models. I’m interested in how post-training can make models reason better and adapt efficiently to new domains.

Before joining Noah's Ark Lab, I was a research assistant at Mila with Professor Aishwarya Agrawal. My work there focused on advancing vision-language models, including fine-grained visio-linguistic understanding, minimal image editing for model refinement, efficient adaptation to new tasks with limited data, and robust automatic evaluation techniques for building more trustworthy AI systems.

I completed my Master’s in Computer Science, specializing in Artificial Intelligence, at Université de Montréal and Mila under Professor Agrawal’s supervision.

Across my work, I’m motivated by building AI systems that are more perceptive, adaptable, and reliable, from multimodal models that understand subtle visual details to language models that reason and adapt more efficiently.

Email  /  Scholar  /  Twitter  /  LinkedIn

Research

My key research interests in the vision-language domain include: fine-grained visio-linguistic reasoning, generative image models, interleaved image-text generation models, robust automatic evaluation methods.