Build a Personal Study Assistant
Turn a stack of lecture notes into a study partner that quizzes you, explains what you got wrong, and tracks which topics you keep failing. You will finish with a reusable system prompt you can point at any subject.
Why this one
Most people paste notes into a chatbot and ask "explain this". That produces a summary you forget in an hour. Active recall — being asked questions you cannot answer — is what actually moves information into long-term memory. This project builds an assistant that does the asking.
Steps
Gather one topic worth of source material
Pick a single topic, not a whole module. Two to four pages of notes is right. Anything larger and the model starts summarising instead of testing, because it cannot hold the detail. Paste the raw notes — do not tidy them first; you want the assistant working from what you actually wrote.
Write the system prompt
The instruction that matters is "ask one question at a time and wait". Without it the model dumps twenty questions and answers in a single reply, which is reading, not recall.
PromptYou are a study partner for the topic below. Ask me ONE question at a time and wait for my answer before continuing. After each answer: say whether it is right, explain the gap in one or two sentences if it is wrong, then ask the next question. Start easy and get harder. Do not summarise the material — I have already read it. Keep a running list of topics I have got wrong and tell me at the end. TOPIC MATERIAL: [paste your notes]
Run one session and note where it drifts
Answer ten questions honestly, including the ones you cannot answer. Watch for two failure modes: questions that are really vocabulary tests rather than understanding, and the model quietly abandoning the one-at-a-time rule after five or six turns. Both are prompt problems, not model problems.
Fix the drift with an explicit rule
Add a correction to your system prompt for each failure you saw. If it asked definition questions, add "prefer questions that require applying the idea to a new example over questions that require recalling a definition". If it lost the one-at-a-time rule, restate it as the final line of the prompt — instructions at the end of a long prompt are followed more reliably than instructions buried in the middle.
Generalise it
Replace your topic-specific wording with placeholders so the same prompt works for any subject. Save it somewhere you will find it again — a note app, a text file, or the PromptForge library. A prompt you cannot find is a prompt you will rewrite from scratch.
You should end up with
A reusable study-partner system prompt plus a list of the topics you failed on.
Done when
- The assistant asks one question at a time for at least fifteen turns without being reminded
- It corrects a wrong answer with an explanation rather than just the right answer
- You can point it at a different subject without rewriting the prompt
If you want to go further
- Add a difficulty ramp: three easy, three medium, then progressively harder
- Ask it to produce a revision schedule from your list of wrong answers
Useful alongside this
Try next
Topics
Related across the site
Ask-Before-Answer for Ambiguous Tasks
For open-ended requests, instruct the model to ask clarifying questions first.
Context engineering
Context engineering is the practice of deciding what information enters a model's context window and in what order. As windows grew, choosing what to include became more consequential than wording the instruction.
Context window
The context window is the maximum number of tokens a model can consider at once — your prompt, any attached documents, the conversation history, and the response it generates. Exceed it and the earliest content gets dropped.
Fine-tuning
Fine-tuning continues training a pre-trained model on your own examples to specialise its behaviour. It is the right tool for teaching consistent style, format or classification behaviour — not for adding knowledge.
Meta-Prompting: Prompts That Write Prompts
The fastest way to a great prompt is asking the model to write and critique one.
Summarise Without Losing the Point
Compress a long document to a tenth of its length while keeping what matters, then verify nothing important was dropped. Tests whether you can direct compression rather than just requesting it.