R&D · lab notebook

Research & notes

Where I write up my own studies and take apart other people's papers. Every claim gets a number, and every number comes with its limitations.

Heading into an M.Sc. in Computer Science · 2027

What I'm curious about

Evaluating AI systems
How do we measure whether a model is actually good, and how far can we trust that measurement?
Information retrieval & search
Helping models find the right knowledge: search, ranking and retrieval-augmented generation.
Efficient & low-resource language models
Smaller, cheaper models and languages with less data: fine-tuning, distillation and running models locally.
Multimodal models
Models that understand more than text: speech, audio and images, and how these signals combine.
Prediction & forecasting
Predictive and time-series models, and the techniques that let us predict further ahead, and more accurately.

Studies I designed, ran and wrote up, each with its code, data and an honest limitations section.

Technical report · Sep 2026RAGInformation retrievalLLM evaluation

Hybrid search is recommended for RAG by default. Does it actually help, and by how much?

context recall, hybrid (vs 0.525 dense): 0.625context recall, hybrid (vs 0.525 dense)
trap questions refused: 9 / 10trap questions refused
of needed evidence in top 5: 17–26%of needed evidence in top 5
retrieval latency, no LLM: 38–83 msretrieval latency, no LLM

Paper · Sep 2026Fine-tuningHebrew NLPRAG

Can a small, open Hebrew model that runs on a laptop learn to answer in plain Hebrew, cite sources and say “I don't know”?

My blog: close readings of papers, lessons from my own experiments, and the road to my M.Sc.