What do you think

What Most People Think AI Is

• A super-intelligent robot that thinks like a human
• Something that will take everyone’s jobs
• A magic box that knows everything
• Dangerous technology from sci-fi movies All of these are wrong. Or at least, incomplete. Let’s talk about what AI actually is.

What AI Actually Is

• AI is pattern recognition at scale.
• You give it enormous amounts of data. It finds patterns. It uses those patterns to make predictions.
• A language model (like Claude or ChatGPT) was trained on billions of pages of text. It learned patterns: what words tend to follow other words, how arguments are structured, how questions are answered.
• It doesn’t "know" anything. It doesn’t "think." It predicts what text should come next based on patterns.
• That prediction is so good it feels like understanding. But it’s not.
• It’s the most powerful pattern machine ever built. That’s impressive enough without calling it intelligent.

What AI Can Do

• Write text — essays, emails, code, poems, business plans
• Analyze text — summarize, translate, extract information
• Generate images — from text descriptions
• Recognize images — identify objects, faces, diseases in medical scans
• Understand speech — transcribe, translate, respond to voice
• Write code — build apps, websites, tools from descriptions
• Reason through problems — math, logic, analysis Two years ago, most of this required a team of engineers. Today, you can do it from your phone, even some watches.

What AI Cannot Do

• Think. It has no understanding, no consciousness, no intentions.
• Be right all the time. It "hallucinates" — confidently generates false information.
• Know what’s true. It predicts plausible text, not factual text.
• Replace human judgment. It’s a tool. You are the thinker.
• Access the internet in real time (most models). Its knowledge has a cutoff. UNLESS YOU ADD SOME TOOLS LIKE EXA.
• Care about you. It has no feelings, no preferences, no agenda. Understanding what it can’t do is more important than knowing what it can.

The Key Mental Model

• AI = Very Good Autocomplete
• What This Means for You
• You type a message
• AI predicts the most likely response
• Based on patterns from billions of examples
• The better your input, the better the output
• It’s completing a pattern, not answering a question This is why prompt engineering matters. You’re shaping the pattern it completes.
• AI is a tool, not an authority
• Always verify important outputs
• The human in the loop (you) is the quality control
• Learning to use AI well is a real skill
• It’s not cheating — it’s a new tool, like a calculator was in 1975 The question isn’t "should I use AI?" It’s "how do I use AI responsibly?"

Why does the same AI

• and garbage other times?
• The difference is almost always the prompt.

The Golden Rule

• Garbage in = garbage out. Gold in = gold out.
• AI completes patterns. Your prompt IS the pattern.
• A vague prompt gives a vague answer. A specific prompt gives a specific answer.
• Bad: "Help me with my essay" → You’ll get generic advice that could apply to anyone.
• Good: "I’m writing a 500-word essay on why I want to study computer science at MIT. My main argument is that growing up in Rwanda without reliable internet made me want to build technology for low-resource environments. Give me feedback on my opening paragraph: [paragraph]" → You’ll get targeted, useful feedback.

The 5 Elements of a Great Prompt

• Role — Tell AI who to be. "You are an experienced university admissions counselor."
• Context — Give background information. "I’m a high school student from Rwanda applying to selective US universities."
• Task — Say exactly what you want. "Review my personal statement and suggest 3 specific improvements."
• Format — Specify how you want the output. "Give your feedback as a numbered list, with the most important first."
• Constraints — Set boundaries. "Keep your response under 200 words. Don’t rewrite my essay — just give feedback."

Advanced Techniques

• Chain of thought: "Think through this step by step before answering." Forces AI to reason, not just predict. Better for math, logic, analysis.
• Few-shot examples: "Here’s an example of what I want: [example]. Now do the same for..." Showing is more powerful than telling.
• Persona stacking: "You are a Rwandan farmer who is skeptical of technology. Respond to this pitch for an AI crop tool." Great for testing ideas, practicing debates, getting diverse perspectives.
• Iterative refinement: Don’t accept the first answer. Say "Make it shorter." "More specific." "Now in Kinyarwanda." Treat it like a conversation, not a search engine.
• Structured output: "Return your answer as a table with columns: Problem, Solution, Cost, Timeline." You control the format. Always.

Prompt Engineering for Building

• This is where it gets powerful. You can build real tools with prompts.
• "Build me a web app that lets users upload a photo of a plant leaf and tells them if the plant is healthy or diseased. Use a simple interface. Make it work on mobile." That’s a prompt. And Claude or Cursor will build it.
• "Create a WhatsApp-style chatbot that answers basic health questions in Kinyarwanda. It should be friendly, simple, and always recommend seeing a doctor for serious symptoms." That’s another prompt. That’s a product.
• This is vibe coding — you describe what you want, AI builds it, you iterate.
• No programming experience needed. You need product thinking: who is this for, what problem does it solve, what should it do.
• Next week, you pick a problem and start building.

What did you hear

The Rules of Problem Selection

• It must be real. Not hypothetical. Someone you know actually has this problem.
• It must be specific. Not "fix education" — "help my classmate practice English pronunciation."
• It must have a user. A specific person or group who will test what you build.
• It must be buildable in one week. You’re shipping v1 next session.
• It doesn’t have to be original. A scholarship finder isn’t new. But one designed for Rwandan students is.

How to Build It

• Open Claude, Cursor, or Replit.
• Describe your tool in plain language. Use the 5 prompt elements from last week.
• Be specific: "Build a web app that does X. The user enters Y. The app returns Z. Make it mobile-friendly."
• Don’t try to build everything at once. Start with the core feature. If it’s a crop disease identifier: just get the photo upload + response working first. If it’s a study scheduler: just get the input form + schedule output working first.
• When something breaks, describe the problem to AI and ask it to fix it.
• Iterate. Version 1 will be ugly. That’s fine. Ugly and working beats pretty and imaginary.

Demo Day Rules

• Every team presents. No exceptions. 5 minutes each.
• Structure your demo: a) The problem — one sentence b) The user — who is this for? c) Live demo — show it working (or show where it broke) d) User feedback — what happened when real people tried it? e) What you learned — what would you change?
• Broken demos are welcome. You learn more from failure than from polish.
• Feedback from the room: what’s strong? what needs work? be specific.
• No slides. Just the product and your voice.

Should students use AI

The Debate

• Yes — It’s a Tool
• No — It’s a Crutch
• Calculators were controversial too. Now they’re required in every math class.
• AI helps students learn faster, not cheat
• The real world uses AI — school should prepare students for reality
• Banning it doesn’t work — students will use it anyway
• The skill is knowing WHEN and HOW to use it, not whether to use it
• If AI writes your essay, you didn’t learn to write
• Struggle is where learning happens — AI removes the struggle
• Students become dependent and can’t think without it
• It’s plagiarism if you submit AI work as your own
• Not everyone has equal access — it creates unfair advantages

A Framework for Responsible Use

• Use AI to learn, not to avoid learning. Good: "Explain this concept to me in simple terms so I understand it" Bad: "Write my essay for me so I don’t have to think about it"
• Use AI as a starting point, not an endpoint. Good: "Give me 3 different angles for my research paper" → then YOU write it Bad: "Write my research paper" → submit as-is
• Always verify. AI makes things up confidently. Check facts, check sources.
• Disclose when appropriate. If AI substantially shaped your work, say so.
• Build your own skills first. Use AI to accelerate, not replace, your thinking.

When AI Gets It Wrong

• AI hallucinates — it generates false information with full confidence. It will cite papers that don’t exist. It will invent statistics. It will get dates wrong.
• AI has biases — trained on internet data, which is biased toward English-speaking, Western perspectives. Ask it about Rwandan history and it may get basic facts wrong.
• AI can be manipulated — with the right prompts, it can be made to produce harmful content.
• The question: who is responsible when AI-generated content causes harm? The company that built it? The person who prompted it? The person who shared it?
• The answer: you. The human in the loop is always responsible.

Rwanda’s AI Policy

• Rwanda has a National AI Policy (MINICT, 2022) — one of the first in Africa.
• It covers 6 priority areas: health, agriculture, urban mobility, finance, governance, education.
• RURA has published "Guidelines on the Ethical Development and Implementation of AI"
• Key principles: transparency, fairness, accountability, privacy, human oversight.
• What this means for you: Rwanda is taking AI seriously at the national level.
• When you build AI tools for your community, you’re part of this story.
• Understanding policy gives you an edge — in applications, in conversations, in your career.

What if the user

• Most AI tools assume a laptop, fast internet, and English.
• What about everyone else?

The Access Problem

• Most AI tools are designed for people with laptops, fast internet, and English fluency.
• That excludes most of the world.
• In Rwanda: smartphone penetration is growing, but many users are voice-first. They call, they send voice notes, they speak — they don’t type long prompts.
• Bandwidth is limited and expensive. Downloading a 5MB response isn’t always viable.
• Kinyarwanda support in major AI tools is minimal.
• This is the gap. And it’s Isomo’s opportunity.
• Voice-first, low-bandwidth, local-language AI is the defensible edge.

Bakame AI

• Bakame is Isomo’s own voice-first AI tool.
• Designed for the Rwandan context: low-bandwidth, Kinyarwanda-capable, voice-driven.
• Use cases: Speaking practice and pronunciation training Debate preparation and interview simulation Information access in Kinyarwanda Tutoring and study support via voice
• This is what makes Isomo’s AI story different from every other EdTech org.
• Today, you’re going to test it and tell us what works and what doesn’t.

Voice AI Design Principles

• Keep responses short. Voice isn’t text — long answers are hard to follow by ear.
• Confirm understanding. "I heard you say ___. Is that right?"
• Handle errors gracefully. "I didn’t catch that. Can you say it again?"
• Be language-aware. Code-switching between Kinyarwanda and English is normal — the AI should handle it.
• Minimize bandwidth. Compress audio. Use short exchanges. Cache common responses.
• Design for noise. Classrooms are loud. Buses are loud. The AI needs to work anyway. If you build a voice-first tool next cycle, these are your design rules.

You shipped v1 last time.

• What broke? What surprised you?
• What would you build differently?

Level Up: What’s Different This Time

• Build Cycle 1 was about proving you can ship. You did.
• Build Cycle 2 is about impact. More users, harder problem, better product.
• You can: a) Improve your v1 — fix what broke, add features users asked for, get to 20+ users b) Start a new, harder project — something with more technical depth or wider reach c) Build a voice-first tool — apply what you learned from the Bakame session
• Solo or team. Your choice. But the bar is higher.
• By next session, you need: a working product, 10+ real users, documented feedback.

Harder Project Ideas

• Crop disease identifier — photo upload, AI diagnosis, Kinyarwanda output. Test with actual farmers.
• Community health bot — WhatsApp-based, answers common health questions, knows when to say "see a doctor."
• Scholarship matching engine — student inputs their profile, gets ranked scholarship recommendations.
• Meeting summarizer — record a voice conversation, get a summary + action items. Test with a community leader.
• Business plan generator — answer 10 questions, get a structured business plan. Test with a young entrepreneur.
• Debate practice AI — pick a topic, AI argues the opposing side, gives feedback on your argumentation.
• Your own idea — something your community told you they need.

The Stakes Are Higher

• Demo Day #1 was about shipping. Did you build something? Good.
• Demo Day #2 is about impact. Did it matter to someone?
• This time, present to a wider audience — other Circles, staff, guests.
• Structure (5 min each):
• The problem — one sentence
• The user — who, specifically
• Live demo — show it working
• The numbers — how many users? what feedback?
• User quotes — what did real people say?
• What you learned and what’s next
• The winning demo isn’t the fanciest. It’s the one with real users and real feedback.

Document Your Project

• After today, write up your project. This becomes a portfolio piece.
• Structure:
• The problem (2–3 sentences)
• What I built (description + screenshots)
• How I built it (tools used, prompts that worked)
• User feedback (quotes, numbers)
• What I learned (honest reflection)
• What’s next (if you kept going)
• This writeup is what you reference in university applications.
• It’s evidence. Not "I’m interested in technology" — "I built a tool used by 30 people."

If you built it but

Why Publishing Matters

• A published piece is a public artifact. Anyone can find it, read it, verify it.
• Admissions officers Google applicants. What do they find when they search your name?
• A published writeup transforms "I’m interested in AI" into "I built X, here’s what I learned, here’s what I think."
• It demonstrates: writing ability, critical thinking, initiative, intellectual depth.
• It’s permanent. It lives beyond this club, beyond this summer.
• And it’s easy. Medium is free. LinkedIn is free. A blog takes 5 minutes to set up.

Three Types of Pieces You Can Write

• Project writeup — document what you built. "I built a crop disease identifier for farmers in my community. Here’s what happened." Structure: problem, solution, how I built it, user feedback, what I learned.
• Opinion piece — take a position on AI. "Why AI literacy should be required in every Rwandan school." "The problem with teaching AI without teaching ethics." Structure: thesis, evidence, counterargument, conclusion.
• Policy analysis — engage with real policy. "Rwanda’s National AI Policy: what it gets right and what’s missing." Structure: summary of policy, analysis, recommendations.

Writing Tips

• Start with a hook. Not "In this article I will discuss..." — that’s dead on arrival. Instead: "Last month, I gave an AI tool to a farmer in my community. He used it to save his tomato crop."
• Write like you talk. Clear, direct, no jargon.
• One idea per paragraph. If it’s longer than 4 sentences, break it up.
• Include specifics. Numbers, names (with permission), quotes, results.
• End with a point. Not a summary — a takeaway. What should the reader think or do?
• Read it out loud before publishing. If it sounds awkward, rewrite it.

What Today Is

• This is not a class presentation. This is a showcase.
• Guests are here: other Circles, Isomo staff, community members, partners.
• You are showing what 10 weeks of building, thinking, and publishing produced.
• Each person/team gets 5–7 minutes:
• Your spike — who are you as a builder?
• Your project — live demo, real users, real impact
• Your published piece — share the link, read the opening
• Your journey — what you learned, what changed, what’s next
• Be proud. You earned this. Most people talk about building things. You actually did it.

What You Walk Away With

• 1–2 built projects with real users and documented impact
• 1 published piece — findable, shareable, permanent
• Prompt engineering and product thinking skills
• Understanding of AI: what it is, what it can do, how to use it responsibly
• Experience with voice AI and low-bandwidth design
• A portfolio that makes you a different kind of applicant When someone asks "tell me about a time you built something" — you have the answer. When an application asks "what are you curious about?" — you have evidence. When a university asks "why should we admit you?" — you have a story no one else has.

What’s Next

• Keep building. The club ends; the skill doesn’t.
• Keep publishing. One piece per month keeps your portfolio alive.
• Keep iterating on your project. Get to 50 users. Then 100.
• Enter competitions. Hackathons, essay contests, fellowship applications.
• Teach someone else. The best way to learn is to teach.
• Stay connected. This cohort built things together — keep collaborating. You started this club 10 weeks ago not knowing what AI was. You’re leaving with a product, a publication, and a skillset. That’s not a club. That’s a launchpad.

You don’t need permission

• You just need a problem, a prompt,
• and the courage to ship it.
• Now go build.