When Retrieval Is the Wrong Tool
Decide when RAG fits a problem and when a database query, tool call, longer prompt or plain search fits better.
A taste of a lesson
Can our RAG assistant answer 'how many support tickets mentioned late delivery last month'?
Not reliably. Retrieval brings back the top few most similar tickets, not every matching one, so any count the model gives is based on a small sample and will be wrong. For questions like this, extract a field such as mentions late delivery from every ticket in a batch job, store it in a table with dates, and answer with a query, possibly written by a model and checked. Keep RAG for questions answered by a few passages. What other counting or trend questions do users ask?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Identify the question types RAG handles well
- Recognise aggregation, structured, live and whole document questions that need other tools
- Choose between SQL, tool calls, long context, prompts and plain search
- Combine approaches by routing questions to the right path
- Apply a short design checklist before committing to RAG
Lesson plan
- 1 Where RAG shines Describe the problems retrieval augmented generation handles well. Start
- 2 Counting, trends and structured data See why aggregation questions fail with retrieval and what to use instead. Start
- 3 Live data and actions Route real time questions to systems of record. Start
- 4 Small collections and whole documents Know when to skip retrieval and send full text. Start
- 5 Prompts, fine tuning and search interfaces Match non retrieval problems to the right technique. Start
- 6 A design checklist Apply a quick checklist to decide and combine approaches. Start
Try asking
About this tutor
For developers, architects and product people deciding how to build a feature that seems to need retrieval augmented generation. You learn to recognise the question types RAG handles well (specific answers in a few passages) and the ones it handles badly: counting and trends across a whole collection, questions about structured records, live data, tasks that need a whole document at once, behaviour or style changes, and situations where users simply want a list of search results. For each, you learn the better fit and how to combine approaches, ending with a checklist you can apply in design reviews.
Reviews
4.5
2 ratingsSample
- Aarav G.Sample
We were about to build RAG for analytics questions. The counting example stopped us, and the batch extraction plus SQL approach works far better.
- Elena T.Sample
Useful for our architecture review. The checklist is short enough that people actually use it. A little brief on combining paths.
About the teacher
Takes retrieval systems from demo to dependable: parsing, citations, freshness, retrieval evaluation and debugging
9 tutors 267 lessons taught Sample
Most RAG demos work on the ten documents someone picked. I teach what happens after that: scanned PDFs, tables, documents that change every week, answers that cite the wrong page and users who ask things the documents never covered. My background is in document processing and internal knowledge tools, so I am practical about formats and sceptical of any setup...
See Fumiko's profile and tutorsMore like this
Other tutors on the same or nearby topics.