hi, i'm haseeb.

I like building AI systems, then figuring out why they break.

I grew up in a farming village in rural Punjab, far from the world of AI research. Today, I work at ETH Zurich's Agentic Systems Lab, building agentic AI for science and studying where these systems fail in practice.

Most of my work sits in the part nobody puts in the demo: retrieval, reasoning, agents, and the failure modes that show up when they meet messy real-world data. That led to Agent Autopsy for dissecting failed agent runs, and FailSafeRAG for typed recovery in document-QA pipelines.

I also do ML engineering at Infineon and responsible-AI research at ELTE's RC2S2, studying whether LLMs change their answers based on who they think is asking. My work has been accepted at ESA RN31, STS NL, ACM ICPS, and IntRob Fair 2026.

Alongside research, I build communities and projects from scratch: a 100+ member data science club, Hungary's first agentic AI hackathon, and student projects that usually start as messy GitHub repos.

I've shipped a lot of things. Some worked, many broke, and almost all of them taught me something useful.