This is my working notebook for AI engineering—small experiments, prototypes, failures, comparisons, and lessons that help turn concepts into something I actually understand.
Building a retrieval-augmented assistant that can answer questions from a private document collection while I learn how chunking, embeddings, retrieval, prompting, and evaluation work together.
How do chunking, embeddings, retrieval quality, prompt construction, and evaluation affect the usefulness of a document assistant?
Which chunking and embedding approaches produce better retrieval quality across different document types and question patterns?
What changes when a model moves from answering questions to using tools, maintaining state, and coordinating multi-step work?
The goal isn’t to collect AI demos. Each experiment should answer a specific technical question, expose assumptions, and leave behind something concrete I can explain or improve.
Define what I’m trying to understand before choosing tools.
A focused experiment is easier to reason about and evaluate.
Useful AI requires more than an answer that merely looks convincing.
What worked, what failed, and what I would test differently next.
New questions will replace old assumptions as I build, test, compare, and understand more.