Evaluating RAG in production
2026-0712 min
Evals show that a RAG system failed. Traces show whether retrieval, context construction, generation, or the eval itself caused it.
Hello!
Most of what I enjoy is working around models, especially on systems that can gather knowledge and produce more accurate answers. That has pulled me into RAG, agentic search, fine-tuning, and agent harnesses, not as a checklist, but as different ways to give models new capabilities and make them useful in specific domains.
Before this, I spent a couple of years across [Public Goods], [DeFi], and fintech, working with Neverland, Giveth, Myosin, and others.
I’m also drawn to where software meets art and human behavior.
Most things we take for granted started with someone being curious enough to look closer.
Notes
Open-weight models
2026-078 min
Open-weights models, proprietary models, parameters, distillation, quantization, inference, deployment, and cost without the usual fog.
NLP basics: NLU, NLG, and core language tasks
2026-067 min
A compact map of NLP, NLU, NLG, tokenization, entity extraction, and sentiment before embeddings enter the picture.
From counting words to embeddings
2026-068 min
How text representation moved from one-hot vectors, Bag of Words, and TF-IDF toward dense embeddings that capture meaning-like similarity.