Five hands-on guides — the rules of prompting, a live practice lab, tool setup walkthroughs, MCP made simple, and the jargon explained with prompts you can try right now.
You don't need all five every time — but when an answer disappoints, one of these is almost always missing. Hover a part to see it in the example.
In the order they most often turn out to be the problem.
Vague asks get generic answers. Name the topic, length and purpose.
Write a post about AI.
Write a 150-word LinkedIn post on 3 AI tools that save students time.
A role shifts vocabulary, depth and what the AI pays attention to.
Help me with my interview.
Act as a hiring manager for a fresher analyst role and run a mock interview.
The AI doesn’t know your situation. One line of context beats three rewrites.
Plan my week.
I have exams Mon & Thu and gym at 7am daily — plan my study week.
One or two examples teach style faster than any description.
Write catchy course titles.
Write 5 titles in this style: “SQL in 10 minutes”, “Excel without fear”.
Tables, bullets, steps or JSON — say it, and the answer becomes usable.
Compare these three laptops.
Compare as a table: price, battery, weight, verdict.
Limits force focus: word counts, reading level, what to leave out.
Summarise this article.
Summarise in 5 bullets, under 12 words each, no jargon.
For big tasks, have the AI interview you before answering.
Fix my business plan.
Before answering, ask me the 3 questions you most need answered.
Steer the draft you have. Point at what to keep and what to change.
Try again.
Keep the structure — make paragraph 2 shorter and add one real example.
Fill in what you can. The prompt assembles itself on the right and the score tells you what is still missing.
Fill in the boxes and your prompt appears here.
Write your own version, then see what a strong one changed.
Write about AI.
Goal: a short explainer you could share with your batch.
You are a tech journalist. Write a 300-word explainer on how AI is changing campus placements in India, for final-year students. Use 3 subheadings and end with 2 action steps they can take this week.
Make my resume better.
Goal: specific, ranked fixes you can act on today.
Act as a recruiter hiring for entry-level data analyst roles. Review my resume below and list the 5 changes that would most improve my shortlist chances, ranked by impact, with a one-line reason each.
Explain machine learning.
Goal: understand it as a non-coder — and check you got it.
Explain machine learning to a commerce student who has never coded, using one everyday analogy, in under 150 words. Then give me 3 quiz questions to check I understood — wait for my answers before revealing them.
Give me business ideas.
Goal: realistic ideas you could start this semester.
Suggest 5 low-cost online business ideas for a college student with ₹10,000 and 10 hours a week. For each give: one-line pitch, how to find the first customer, and the biggest risk. Format as a table.
52 tools worth knowing, grouped by the job you are doing.
MCP is how an assistant reaches your files, your repositories or your database — through one common connection instead of a different integration for each.
search_files({ path: "~/Desktop", pattern: "*.pdf" })list_issues({ repo: "you/portfolio", labels: ["bug"] })search({ query: "resume" })query({ sql: "SELECT course, count(*) FROM enrolments WHERE …" })Somebody has already written one for Drive, Slack, Postgres and a hundred other things.
Your assistant has a settings file for this. One block of JSON.
Name it, point it at the folder or account, restart the app.
You choose what it may see. It cannot reach anything you did not hand it.
“What changed in the spec this week?” It goes and looks.
Each one in plain English, with the picture that makes it stick.
A Large Language Model is an AI trained on huge amounts of text to predict the next word. At scale, that one skill lets it write, summarise, translate and reason.
Models read text in chunks called tokens — roughly three-quarters of an English word. Limits and pricing are counted in tokens, not words.
The maximum amount of text a model can consider at once — your prompt, uploaded files and the conversation so far.
A setting that controls randomness. Low gives focused, repeatable answers; high gives creative, varied ones.
Models that understand more than text — images, audio, PDFs, screenshots — and sometimes generate them too.
When a model states something false with confidence — invented stats, citations or facts.
Hidden instructions inside a web page or document that hijack what the assistant does next. The reason an agent should not be trusted with everything.
The standing instruction an app gives the model before you type anything — its rules, its voice and what it must refuse.
Giving a few examples of input → output so the model copies the pattern exactly.
Asking the model to reason step by step before giving a final answer. It improves accuracy on maths, logic and planning.
Lists of numbers that capture the meaning of text, so a computer can find similar ideas — not just matching words.
Retrieval-Augmented Generation: the system first finds relevant passages in your documents, then the model answers using them.
Further training a model on your own examples so a style or skill is built in. Usually only worth it after prompting and RAG fall short.
A standalone piece of content the AI creates beside the chat — a document, code file, chart or mini app — that you can view, edit and reuse.
Also called function calling: the model calls external functions — web search, calculator, calendar, code — and uses the results.
An AI that plans steps and uses tools in a loop to reach a goal, checking its own progress along the way.
Model Context Protocol: an open standard that lets AI apps connect to tools and data — files, GitHub, databases — through one common plug.
This page is the first week. The courses are the rest — with someone to ask when a prompt will not behave.