Research

Experiments that ask a narrower question, then test it hard.

Long-form technical notes on retrieval, context efficiency, tabular foundation models, and evaluation under real constraints.

Retrieval · Multi-hop RAG

Context-Efficient RAG for Multi-Hop Reasoning

Does reasoning-aware retrieval improve evidence coverage without paying for unnecessary context? I compare reasoning-unit retrieval with conventional chunks across retrieval quality, context efficiency, evidence coverage, and answer quality.

Read research →GitHub ↗

Tabular ML · Foundation Models

Nori Benchmark

Can an in-context tabular foundation model preserve an advantage as context grows, and how robust is it to irrelevant numerical features? Multi-seed experiments on NYC Street Tree Census data.

Read benchmark →GitHub ↗