Interactive learning session 1 — Foundations
Welcome to AMPLIFY. Over four months you will take one real problem from your own work and build a working AI-assisted solution to it — ending with a group presentation. This first class gets you set up on the tools, shows you how the course runs, and teaches the habit everything else is built on: writing prompts that get you what you actually want.
This page is part of the course. Everything here is interactive — switch languages at the top, tick the setup checklist, pass the three skill checks, and build your first prompt. Your progress is saved in your browser only.
1 · How the course runs
Four months, eight classes, one growing project. Know the rules of the road before we start.
Homework is not optional — it is how you stay in the course
A homework assignment follows every class. Each one practices the topics we just covered and applies them to your own project. Every class opens with a student presenting their solution. You will all present during the course — homework is your ticket to the showcase.
Tools we leave at the door: Email WhatsApp Slack only
Set up before Class 1
Tick as you go — progress is saved on this device and shows in the ring at the top.
2 · Slack — where the course lives
Slack does three jobs in this course: support, community, and homework. Learn your way around and you will never wait for an email reply.
Your map of the workspace
House rules: post publicly by default (your question helps a colleague), reply in threads, both languages are welcome everywhere, and @mention only when you need someone specific.
How homework flows
- Get itThe assignment is posted in the module channel after class.
- Do itWork on it in your Claude project, on your own problem.
- Submit itUpload the file to your private channel, titled FirstNameLastName_Assignment1.
- Get feedbackInstructors reply in a thread — feedback stays attached to your work, not lost in a chat.
3 · Claude — your AI teammate
Every student gets a Claude Teams account. Today you only need three things: install, log in, and find Projects. In Class 2 we compare Chat, Cowork, and Code (the CLI) — the right tool for the job.
Accept the Teams invite
Your instructor sent you an invite to the AMPLIFY Claude Teams plan. Accept it with the email you gave us — that account is your course login.
Install the desktop app
Get the app for macOS or Windows from claude.ai/download. Mobile apps exist too — install whatever you like.
Log in and look around
Open the app, log in with your Teams account, and spend ten minutes clicking around. The map below shows you what matters.
The project rule: do your project work inside your project
We create one project for every student. Open it from Projects in the sidebar and start there whenever you touch your course problem — your files, context, and best prompts stay together, so Claude gets better every week. Working on something else? The main workspace is fine.
Find your way around — click the pulsing dots
(Illustration — the real app looks a little different, but the parts are these.)
Coming in Class 2 — the right tool for the job
Ask, think, draft. A conversation in a window — what you used today.
Hand Claude a whole task: it works in your files and apps, then reports back.
Claude works directly in your terminal and project files — power tools for real builds.
For today: be logged in, find Projects, poke around. Familiar beats perfect — we go deep next class.
Red lines — what never goes into Claude (or any public AI tool)
- Patient names, phone numbers, or registers that identify people
- Passwords, API keys, or credentials of any kind
- Unpublished strategy documents or confidential drafts
Safe: published guidelines, public programme data, anonymised aggregate numbers, your own drafts.
4 · The art of the perfect prompt
Before any tool matters, this matters. AI does not read your mind — vague in, vague out. Think of Claude as a brilliant new colleague who has never seen your programme: it needs direction, context, and a clear definition of "done".
Why prompting is the whole game
A large language model doesn't "know" your districts, your template versions, or your deadline. It predicts the most plausible next words given what you gave it. When you leave a gap — no audience, no format, no check — it fills it with a confident guess. That's where hallucinations and generic waffle come from.
Prompting is specifying just enough reality that the model doesn't need to invent. Persona tells it whose judgment to use; Task names the verb; Context gives the facts it cannot know; Format says what "done" looks like; Iterate is your safety net — the check you own, not the model.
"As an NTD M&E officer" is not flattery — it steers tone, vocabulary, and what to prioritise (coverage thresholds, not generic advice).
Audience, purpose, constraints, the data you actually have. The model never has these unless you give them.
The only reliable anti-hallucination step is yours: "flag anything over 100%; ask me before finalising." Build it into the prompt.
Persona
Who is asking? Give the AI an expert perspective to think from.
"You are an NTD monitoring & evaluation officer…"
Task
What exactly should be done? One clear verb: summarize, compare, draft, check.
"Summarize coverage by district for 2025."
Context
Why and for whom? Audience, purpose, constraints, what data you have.
"For the national programme review; aggregate district data only."
Format
How should it look? Table, bullets, one page, formal French…
"A table plus three bullets, formal."
Iterate
Was it right? Evaluate, refine, run again. Good prompting is a loop, not a wish.
"Flag coverage over 100%; ask me before finalizing."
Six prompt types you'll meet (beyond P-T-C-F-I)
P-T-C-F-I is how you compose. These are flavours you'll recognise on Slack, GitHub, and in the wild — most are just P-T-C-F-I done briefly. Spot which you already use.
| Type | When to use | Example |
|---|---|---|
| Zero-shot | Simple instruction, no examples. Good for one-off summaries. | "Summarize this article in 5 bullets." |
| Few-shot Try this | Give 2–3 examples of the form/tone you want, then ask for the next in the same style. Your best shortcut to consistency. | "Here are 2 example district summaries. Write a third in the same style: [District A: ...] [District B: ...]" |
| Instructional | Direct verb + constraints (write / compare / keep under X words). | "Write an executive summary. Keep it under 100 words, formal French." |
| Role-based | Persona on purpose — "as an M&E officer..." steers lens and vocabulary. | "You are an MBA professor preparing a lecture outline..." |
| Contextual | Background + audience before the ask. That is your C. | "This is for community volunteers. Rephrase in simple language." |
| System / Meta | Behind-the-scenes rules for tone/scope ("always cite sources, never guess"). You won't write these in Class 1 — but you'll recognise them in skills. | "Always respond formally and cite real sources. Never guess." |
A worked example — colour-coded
Spot the parts: persona → task → context → format → iterate. Still room to improve — "easy-to-read format" is vague. Say what columns you want.
Same request, two prompts — see the difference
Lab — watch the output sharpen as you add detail
Start vague. Layer P-T-C-F-I one tap at a time and watch the mock output go from generic to usable. (No AI call — just the mechanics.)
How can AI help my NTD programme?
Three phrases worth memorizing
- "Think through this with me step by step."
- "That's helpful, but I need it specific to my context: …"
- "Ask me questions to clarify my goal and surface my assumptions — don't decide for me."
Autopsy — three real prompts from the field
Open a case, spot what's missing, check your eye against the coach's. No typing — just diagnosis.
Prompt builder — write your first structured prompt
Fill the five parts, watch your prompt assemble, then copy it into Claude and iterate. Your real task beats any example.
Try it on any prompt — paste what you'd actually send
Not using the builder? Paste a real prompt from your work (or draft one from scratch) and get the same P-T-C-F-I diagnosis. AI-backed when online; heuristic offline.
Suggested rewrite
Could we build a skill for the perfect prompt?
Yes — and that is exactly the plan. A skill is a saved, reusable set of instructions you can call on demand instead of retyping. The builder above is the draft of that skill: the P-T-C-F-I anatomy plus your house rules. Today you learn it by hand.
In Class 2 we turn your best prompts into a real skill, hunt for existing skills (they are shared, for example, on GitHub), and you learn to tell reusable prompts from one-offs. Homework prep: bring one task you repeat every week — that is your first skill candidate.
5 · Your problem — a starting draft
You don't need a perfect problem statement to start. You need something real enough to try today's prompting on — and honest enough to improve. Every class in AMPLIFY sharpens the same problem, so today's draft is the raw material, not the final answer.
It sharpens every class — you don't have to get it all now
Class 1 you name it. Class 2 you learn to prompt it. Classes 3–4 you connect it to your real data. Classes 5–7 you build the checks, the dashboard, and the pitch. Class 8 you present. Each layer reveals what you were missing — so we keep the problem statement draft and revisit it. Perfection now would just hide the next lesson.
Find your problem in two prompts
About 20 minutes. Run the two prompts in order, in the same Claude chat, and record what you find in the boxes here. Your answers save in this browser as you type. At the end, download your problem file.
Let the AI interview you about 15 min
This gets you talking about your actual work before you jump to a solution. Fill in the three boxes — they go straight into the prompt.
I work as a [role] on [programme/disease] in [country/district]. I’m starting a course where I’ll use AI to work on a real problem from my job, but I don’t know what that problem is yet. Interview me to help me find it. Ask me one question at a time about my normal working week: tasks that are repetitive, slow, frustrating, or that I keep putting off. Don’t suggest solutions or AI tools. After about 6 questions, stop. Then write a short summary I can copy of what the interview brought up, and name the 2–3 tasks that came up most.
Turn the answers into options about 5 min
Based on my answers, describe 3 possible problems I could work on. For each one, write one plain sentence: who struggles, with which task, and why it matters. Don’t mention AI or technology. Then ask me which one feels most real to me.
Write your problem statement about 5 min
Using the option I chose, write my problem statement in two sentences: who struggles, with which task, and why it matters. Write it plainly, in first person, without mentioning AI or technology. Then ask me if it sounds like my work.
Download your problem file
Your instructors review this file and give you feedback. Keep it safe: in session 2 you upload it, and the next section builds on the option you chose.
- Step 1 · Interview not finished
- Step 2 · Options not finished
- Step 3 · Statement not finished
- Full page · all sections read not finished
Some steps aren't finished yet — the file will show them as incomplete.
Saves a web page (.html). Open it in any browser to check it, then post it on Slack as your homework.
Homework — due before Class 2
Four steps, all on Slack. Class 2 opens with one of you presenting.
Introduce yourself on #all-amplify
Name, role, country — and one sentence on the problem you are working on.
Post your problem
Paste the option you chose.
Reply to two classmates
One thing you find interesting about their problem, and one question back.
Run the prompt builder on a real task
Pick something you actually did this week. Build the prompt, run it in Claude (inside your project), and note what changed.
Submissions go to your private channel, titled FirstNameLastName_Assignment1. Homework is required to stay in the course.
Quick answers
What if I miss a live class?
Watch the recording in the module channel, do the homework anyway, and ask on Slack. Don't disappear — the course moves every two weeks.
What if Claude gives me something wrong?
Treat it as a prompt problem first: iterate (the I in P-T-C-F-I is you). Verify every number against a source. If it keeps happening, post the prompt and output on Slack — debugging prompts is a group sport.
Can I work in French?
Yes. Write to Claude and on Slack in whichever language you think in; instructors read both. (This page toggles with EN/FR at the top.)
Where do I find everything?
One place: Slack. Slides, templates, recordings, and homework feedback all live there — this page is the only exception.