Research notes and what's been on my mind.
How Modern Recommendation Systems Recommend
Explaining how the recommendation systems we interact with everyday know exactly what it takes to keep you on their platform.
MLEmbeddingsTwo Tower ModelsVector SearchRanking
Scalable RAG Evaluation Without a Golden Set
A compact note on judging retrieval quality with lightweight probes, stratified slices, and model-assisted review.
RAGEvaluationRetrieval
Photographing Vector Spaces
A visual framing for embedding neighborhoods, failure clusters, and the stories hidden in nearest-neighbor graphs.
EmbeddingsVisualizationComputer Vision