Farid Karimli

I’m a first-year PhD student in Data Science in the Faculty of Computing & Data Sciences at Boston University. I study the design of proactive, multimodal assistants that transcend individual devices to support humans in situated, daily tasks. I plan to bridge low-cost computer vision and agentic ML to build systems that don’t just react to commands, but anticipate user needs by reasoning over their environment and past interactions.

My work sits at the intersection of AI, HCI, and software — spanning efficient video understanding, benchmarking proactiveness in multimodal models, and the design of new interactive experiences powered by agentic AI. I have a particular interest in quantifying the “intervention window”: the moment where AI assistance adds the most value, before the user has to ask.

My background is in computer science and data science. Previously, I developed educational AI tutors for ML courses at BU, accessibility tools including CameraMouse and KeyGlide, and AI agents for workplace productivity in sales. I have published in accessibility and HCI, and bring a strong foundation in data processing and full-stack web development.

Current Work: I am currently pursuing two parallel threads. The first is a deeper investigation into vision model behavior under extreme label cardinalities in the fine-grained setting — a foundational effort to understand how vision models scale and where they break down. The second is DeepRead, an application to bridge the gap between academic papers and their code implementations, rendering both in a single unified view to make research more navigable for practitioners and researchers alike.

I am always open to conversations and collaborations at the intersection of AI and HCI — feel free to reach out.