Getting the right information in front of the model at the right time, and making the answer checkable afterwards. Most quality problems that look like model problems are context problems, and most context problems are product decisions wearing an engineering costume.
| # | Lesson | The question it answers |
|---|---|---|
| 01 | Context Engineering | What goes in the window, what it costs, and what gets lost in the middle? |
| 02 | Retrieval and Grounding | What does retrieval actually fix, and what does it not? |
| 03 | Chunking, Ranking and Recall | Which retrieval decisions are product decisions? |
| 04 | Citations and Provenance | How do you make an answer checkable by the person reading it? |
| 05 | Measuring Retrieval Separately | Why must recall and faithfulness be measured apart? |
| 06 | Knowledge That Goes Stale | What is the half life of a wrong answer, and how do you invalidate? |
| 07 | Memory and Personalisation | Session, user and organisation memory as a product surface |
| 08 | Permissions Inside Retrieval | How do you stop the model citing something this user may not see? |
| 09 | Structured Data and Tools Instead of Text | When is a query better than a document? |
| 10 | When Finetuning Is the Answer | Rarely. What do you do instead, and what is the exception? |
| 11 | The Context Budget as a Product Decision | Who decides how many documents, and on what basis? |
| 12 | Worked: A Grounded Assistant | One real retrieval product, from corpus to citation |