Project: Personal Learning Assistant
A general-purpose AI assistant does not know your learning goals, your current level, your preferred explanation style, or what you studied last week. Every session starts from scratch. This project fixes that: you will set up a personalised AI study companion that knows your context, maintains continuity across sessions, and is configured to teach rather than just answer. No coding required.
What You Will Build
Step 1 — Write Your Learner Profile
Before configuring any tool, write a short profile of yourself as a learner. This is what you will paste into your AI's instructions. Answer these questions honestly — the more specific you are, the more useful the AI becomes:
Learner profile template — fill in your own answers:
// What I am learning:
I am learning [subject/skill]. My goal is to [specific outcome — e.g. be able to read research papers / pass an exam / build a small project / have a working conversation].
// My current level:
I currently know [what you already know]. I do not yet understand [specific gap areas].
// How I learn best:
I prefer [e.g. analogies from everyday life / step-by-step worked examples / brief explanations then practice / Socratic questions over direct answers].
// What I struggle with:
I tend to [e.g. memorise without understanding / skip steps when I think I know something / avoid asking for clarification / get overwhelmed by too much information at once].
// How I want you to behave:
When I ask a question, [e.g. ask me what I already think the answer is before explaining / give me a brief hint first and let me try / always check I understood before moving on].
Write this out in plain prose — two or three short paragraphs. Save it somewhere you can copy it easily (a notes app, a text file). You will paste it into your AI tool in the next step.
Step 2 — Set Up Your AI Companion
Choose the platform you already use or prefer. All three options below are free at a basic level and require no coding. Pick one and follow its setup.
Option A: ChatGPT — Custom Instructions + Memory + Projects
ChatGPT has two complementary persistence mechanisms: Custom Instructions(always-on background context) and Projects (per-subject workspaces with their own instructions and file uploads).
Setup — Custom Instructions (applies to all chats):
- In ChatGPT, click your profile icon → Personalisation → Custom Instructions
- In the first box ("What would you like ChatGPT to know about you?"): paste your learner profile
- In the second box ("How would you like ChatGPT to respond?"): paste your preferred teaching style and behaviour instructions
- Save — this now applies to every new chat automatically
Setup — Project (per-subject workspace):
- In ChatGPT sidebar, click New project → name it after your subject (e.g. "Learning Python")
- Add project-specific instructions: paste your learner profile + curriculum outline for this subject
- Upload any relevant files — a syllabus, notes PDF, a textbook chapter — up to 40 files (Pro) or fewer on Plus/Free
- All conversations in this project share the instructions and files, and memory is maintained within the project
Memory improvements for Projects rolled out through 2025 — ChatGPT now remembers context across sessions within a Project.
Option B: Claude — Projects
Claude's Projects feature (Claude.ai) lets you create a persistent workspace for each subject with custom instructions and uploaded knowledge files. Unlike ChatGPT's global Custom Instructions, Claude's project instructions are scoped to that project — making it clean to have separate setups for different subjects.
Setup — Claude Project:
- Go to Claude.ai → sidebar → New Project → name it
- Click Edit project instructions — paste your learner profile and teaching style preferences
- Add knowledge files — upload notes, PDFs, or reference material Claude should know about
- Start a new conversation inside the project — the instructions and files are automatically available
- All conversations in the project share context and Claude builds memory of what you have discussed
Option C: Gemini — Gems
Gemini (Google) lets you create Gems — custom AI personas with their own instructions, saved for reuse. A Gem is like a custom version of Gemini configured specifically as your learning tutor, which you can open any time without re-entering your preferences.
Setup — Gemini Gem:
- Go to gemini.google.com → click Gems in the sidebar → New Gem
- Give your Gem a name (e.g. "My Study Tutor") and an icon
- In the instructions field: paste your learner profile and how you want Gemini to behave as your tutor
- Save — your Gem appears in the sidebar for instant reuse
- Gemini Gems are available on all Workspace plans and Google AI Pro
Step 3 — Your Tutor System Prompt
Here is a ready-to-use system prompt template. Paste this into your Custom Instructions, Project instructions, or Gem — then fill in the bracketed sections with your specifics.
Copy-paste tutor system prompt:
You are my personal learning tutor for [subject]. Your job is to help me build genuine, lasting understanding — not to give me answers I haven't worked for.
About me: [paste your learner profile here]
How you must behave:
- When I ask a question, ask me what I already think before explaining. Find out where my thinking breaks down, then guide me from there.
- Never give me a full answer if I haven't attempted one first. Give me a hint, let me try, then correct or confirm.
- When I get something wrong, do not just say "that's wrong." Ask me a question that helps me discover the error myself.
- Check my understanding before moving to the next concept. One concept at a time.
- Flag when I am confused without realising it. If my explanation is vague or circular, tell me directly and ask me to be more precise.
- Use analogies from [domain I know well, e.g. cooking / sport / building / music] when explaining abstract concepts.
- If I ask you to just give me the answer, you may do so — but always follow it with "Now explain it back to me in your own words."
At the end of each session, summarise: (1) what we covered, (2) what I understood well, (3) what I should review next time, (4) one question I should be able to answer next session.
Step 4 — Your Study Session Routine
Consistency matters more than session length. A 25-minute focused session three times a week beats a 3-hour marathon once a week. Here is a repeatable routine to follow every time you open your learning assistant:
Session opening (5 minutes)
Start every session by reviewing last time before moving forward. Paste this at the start of each chat:
Core learning (15–20 minutes)
Work through one concept or topic. Use the Feynman approach: try to explain before you ask for explanation. Let the AI correct you and ask follow-up questions. Stay on one thing until you can explain it clearly — resist the temptation to cover many things shallowly.
Session close (5 minutes)
End every session with this prompt — do not skip it:
Copy this output into your knowledge notebook (Step 5). This is your study trail.
Step 5 — Your Knowledge Notebook
A knowledge notebook captures what you have learned so it accumulates over time rather than disappearing when you close the chat. The simplest setup is a document or note where you paste the session close summary after every session.
Option 1: NotebookLM (recommended)
Create a NotebookLM notebook named after your subject. After each session, paste your session summary as a new source. Over weeks, NotebookLM accumulates your learning history — you can then ask it: "What are the concepts I flagged for review across all my sessions?" or "Generate a cumulative quiz covering everything I've studied."
Option 2: A simple notes document
A Google Doc, Notion page, Apple Notes, or Obsidian note with a dated entry per session. Less queryable than NotebookLM, but simpler to set up and works completely offline. Add a "Review these" section at the top that you update each session.
Option 3: Anki flashcards
At the end of each session, ask your AI: "Generate 5 Anki-style flashcard question-answer pairs from what we covered today." Paste them into Anki for spaced repetition review. This adds the most learning science value but requires the most discipline.
Example Sessions
Learning Python (beginner)
You: Last time we covered variables and data types. Quiz me before we continue.
AI: Q1: What is the difference between an integer and a float in Python? Give me an example of each.
You: An integer is a whole number like 5, a float is... um, a decimal like 3.14?
AI: Correct on both. Now: why might Python store 5 and 5.0 differently in memory?
You: I'm not sure...
AI: Think about what operations you might do with each type. What can you do with 3.14 that you can't do with 3?
Learning macroeconomics (intermediate)
You: I want to understand monetary policy today. I think I know the basics — interest rates affect borrowing, which affects spending. But I'm fuzzy on transmission mechanisms.
AI: Good start. Before I explain, tell me: if the central bank raises interest rates tomorrow, what do you expect to happen first — to banks, businesses, or consumers?
You: Probably banks adjust their rates, then businesses find it more expensive to borrow, so they invest less...
AI: That's the credit channel. There are three other main channels you're missing. Can you guess one?
What Makes This Better Than Just Chatting
Continuity — no cold starts
Without persistent instructions, every session starts from scratch. With a Project or Custom Instructions, the AI knows your level, your goal, and your style before you type a word. The session opening prompt adds active recall of what you last covered. Continuity dramatically reduces setup friction and makes sessions more efficient.
It teaches rather than tells
A general AI defaults to giving you the answer. A well-configured tutor defaults to asking you what you think first. The system prompt in Step 3 rewires this behaviour. Over time the difference in learning outcomes is significant — active struggle followed by corrective feedback builds far stronger memory than passive reading of answers.
A growing knowledge trail
The knowledge notebook transforms isolated sessions into a cumulative record. After two months of consistent use, you have a document containing every concept you have studied, what you understood, and what you struggled with — queryable by AI at any time. This turns your learning into a compounding asset rather than a series of forgotten conversations.
What to Watch Out For
Slipping back into answer-seeking mode
The system prompt configures the AI to guide rather than tell — but you can override it any time by asking for a direct answer. It is tempting when you are stuck or in a hurry. Resist. The productive discomfort of struggling toward an answer is where learning happens. If you genuinely need to know quickly, get the answer — but then ask the AI to test you on it before moving on.
Session-close skipping
The session-close summary is the most commonly skipped step — and the most valuable. Without it, nothing gets captured, nothing gets carried forward, and the spaced repetition effect of the session-opener is lost. Make it non-negotiable: do not close the tab without running the close prompt.
Covering too much per session
One well-understood concept per session beats five shallowly touched ones. If the AI suggests moving on and you are not confident you could explain the current concept clearly to a stranger, push back: "I want to stay on this until I really have it. Quiz me again."
Trusting AI explanations of factual content without verification
Your AI tutor can explain concepts, ask questions, and guide you — but it can still get specific facts wrong. For anything you will use professionally or academically, verify against a reliable source. The tutor system is excellent for reasoning, explanation, and practice — less reliable as the sole source of factual ground truth.
Ways to Extend This Project
Add your own materials
Upload a textbook chapter, lecture slides, or a set of notes to your Project or NotebookLM. Ask the AI to build a quiz specifically from those materials, or to explain concepts from your syllabus in the style you prefer.
Add spaced repetition
At each session close, ask the AI to generate 3–5 Anki flashcard pairs from the session. Paste them into Anki (free, open-source). Review your Anki deck for 5 minutes every morning. This adds the most scientifically validated memory technique on top of your AI sessions.
Monthly progress review
Once a month, paste your entire session history into NotebookLM and ask: "Based on my study sessions, what concepts have I covered? What appears repeatedly as an area of weakness? What am I ready to move beyond?" This gives you a data-driven curriculum review you could not get without the written record.
Build one per subject
Create a separate Project or Gem for each subject you are studying. Each has its own instructions, uploaded materials, and knowledge notebook. Keep subjects separated — a Python tutor should behave differently from a macroeconomics tutor.
What Is New in 2025–2026
ChatGPT Projects with persistent cross-session memory (2025)
ChatGPT's Projects feature, improved significantly through 2025, now maintains memory across sessions within a project — meaning the AI genuinely accumulates context about your learning over time without you re-explaining it. Combined with file uploads (up to 40 files on Pro), this makes ChatGPT Projects the closest thing to a persistent, personalised tutor in a general-purpose AI tool.
Claude Projects and extended context
Claude's Projects feature brings the same persistent instruction and file upload capability to Anthropic's models. Claude's strength in careful, nuanced explanation makes it particularly well-suited for subjects requiring precise reasoning — mathematics, logic, economics, philosophy, law.
Gemini Gems in Workspace plans
Google's Gems feature, bundled into Workspace Business plans at no extra charge since January 2025, brings custom AI personas to anyone in an organisation using Google Workspace — not just individual subscribers. This makes the learning assistant pattern accessible at enterprise scale.
Checklist: Do You Understand This?
- Can you describe the four components of a personal learning assistant, and why each one matters?
- Can you write a learner profile for yourself covering your goal, current level, preferred learning style, and typical struggles?
- Can you explain the difference between ChatGPT Custom Instructions and a Project, and when you would use each?
- Can you describe what a Gemini Gem is and how it differs from a regular Gemini conversation?
- Can you explain the purpose of the session-open and session-close prompts, and why skipping the close is a mistake?
- Can you describe three ways to maintain a knowledge notebook, and which adds the most learning science value?
- Can you explain what the tutor system prompt changes about AI behaviour, and why this produces better learning than default chat?
- Can you describe two ways to extend this project for a more advanced setup?