Agentic AI Systems
Designing autonomous agent architectures that plan, execute, and verify complex multi-step tasks with explicit evidence trails.
Advancing computer science through applied AI research.
We investigate the frontiers of agentic systems and human cognition — how AI agents reason, remember, orchestrate tools, and use language, and how engineers think, delegate, and adapt when working alongside them.
Our work sits at the intersection of AI systems engineering, model architectures, natural language, and cognitive science — building tools and studying how they change the way engineers think and collaborate.
Designing autonomous agent architectures that plan, execute, and verify complex multi-step tasks with explicit evidence trails.
Studying transformer variants, inference optimization, and model selection strategies that power modern agent systems.
Designing protocols and runtime layers that let agents discover, compose, and invoke external tools reliably at scale.
Investigating how language models parse, generate, and reason over natural language — and what that reveals about communication itself.
Studying how engineers think, delegate, and adapt when working alongside AI — the cognitive and behavioral dimension of human–machine collaboration.
Researching long-term context retention across sessions — decisions, preferences, and project knowledge that survive restarts.
Studying how developers delegate reasoning to AI systems and designing interfaces that augment human cognition.
Conference talks, publications, and written research from the lab.
An exploration of how large language models shape representation versus reality, the intersection of NLP and human cognition, and what critical thinking means in the age of responsible AI usage.
Read paper →Consi is an open-source cognitive AI runtime for developers — a self-hosted platform that combines persistent memory, agent orchestration, modular skills, and native tools into a single engineering workbench. It is the primary testbed for our research on agentic systems and cognitive offloading.
Learn more about ConsiExperimentation and studies with Serverless Architectures and Data Aggregations
Follow World Cup liveInterested in our research or want to contribute? Reach out to the lab directly.