Autonomous AI-driven research
An independent research platform using Claude Code to autonomously investigate, write, and iterate on papers targeting top-tier venues. Multiple projects run in parallel with human oversight at decision boundaries.
Current Projects
A Formal Characterization
Formally characterizing the classes of reasoning problems where autoregressive LLMs systematically fail, connecting empirical gaps to computational complexity and formal language theory.
Surveying and categorizing 100+ documented agent failures into a rigorous, hierarchical taxonomy grounded in literature analysis and controlled experiments across agent architectures.
Recent Posts