AI Student Project Starter Pack

You commented PROJECT. Here are 3 beginner AI projects you can build on a student budget, put on GitHub, and talk about with confidence in your next internship interview. After those you'll find 9 more ideas.

Knowing ChatGPT is not enough

Everyone's CV says "proficient in ChatGPT", so it doesn't make you stand out. This does: "I saw a problem, I built an AI solution, here's the link, and here's what I learned when it broke."

That's what I saw at the Mpumalanga AI Student Hackathon. Students from Tshwane University of Technology and the University of Mpumalanga didn't just talk about AI. They built things:

• SignSure: identity verification for digital signatures.
• AgricShield AI: helping agriculture recover from market disruptions.
• MineLink & MineGuard: mining safety.

You don't need to be an AI engineer to start. You need one small problem, one weekend, and this loop:

1. Spot a problem that you, your friends or your community actually have.
2. Build small. Version 1 does one thing well. Ugly is fine.
3. Put it on GitHub. That's your public proof of work.
4. Document it: what you built, what went wrong, what you learned.
5. Use it in internship interviews and on LinkedIn.

Tip: Start small. A tiny tool that works beats a huge idea that never ships. Every project here has a Version 1 you can finish in a weekend.

Tools, free tiers and links were checked on 30 September 2026. Tech changes fast. If something looks different, check the official page and current pricing.

Before you start: your free toolkit

You can build everything here for R0 in software costs. What you need is a laptop or PC with a browser, a Google account, and data or Wi-Fi.

Tool	Use it for	Cost (checked 30 Sep 2026)
GitHub	Storing your code and README (your portfolio)	Free plan with unlimited public and private repos. Students can apply for the Student Developer Pack, which includes free GitHub Pro while you're a student.
Google AI Studio	Testing prompts without code, and getting your Gemini API key	Free tier on selected models, with rate limits. Google says it uses free-tier content to improve its products, so don't paste private info.
Groq	Backup LLM API (open models)	Free tier with rate limits. Check current limits.
Google Colab	Running Python in your browser with no install	Free, but resources aren't guaranteed. The AI features need an 18+ account.
Streamlit and Community Cloud	Turning Python into a web app and deploying it from GitHub	Open-source. Community Cloud hosting is free.
Gradio	Quick AI demo screens	Free, open-source library.
Hugging Face Spaces	Another place to host demos	Static Spaces are free. Most Gradio/Docker Spaces now need the paid PRO plan, though free accounts can host up to 2 Gradio Spaces on ZeroGPU. Check current pricing.
VS Code	Code editor	Free.
Cursor	AI code editor that can explain and fix your code	Free Hobby plan with limited agent requests and no credit card.
Gemini Notebook (formerly NotebookLM)	No-code study-assistant prototype	Free standard tier with daily limits.
Python	Programming language for the code path	Free.

Tip: Save data. Download big installers on campus Wi-Fi. An AI API call only sends text, so it uses very little data compared with video. Skip big AI models that run on your own laptop for now, because the downloads are huge. Close Colab tabs when you're done.

Warning: Keep your API key secret. Treat it like a PIN and never push it to GitHub. Keep it in Streamlit Secrets or Colab's Secrets panel instead. If it leaks, delete that key in AI Studio and create a new one.

Choose your path, then use the starter code

No-code route. Write and test your tool's instructions (the system prompt) in Google AI Studio. AI Studio's Build mode can generate a simple app from a description, and eligible accounts can publish up to 2 apps free on Google's Starter Tier. This route is good for testing an idea fast, but interviewers will ask what's under the hood.

Code route. Use Python, Streamlit and the Gemini API, pushed to GitHub and deployed free on Streamlit Community Cloud. Version 1 of each project takes about 20–60 lines. This route is the better one for your portfolio.

Starter skeleton

Reuse this for every project.

# app.py
import streamlit as st
from google import genai

client = genai.Client(api_key=st.secrets["GEMINI_API_KEY"])  # key lives in Secrets, not in code

RULES = """You are a helpful assistant for South African students.
Be clear, friendly and honest. If you are not sure, say so."""

st.title("My first AI tool")
user_text = st.text_area("Type or paste something")

if st.button("Go") and user_text:
    with st.spinner("Thinking..."):
        result = client.interactions.create(
            model="gemini-3.8-flash",  # check ai.google.dev for current free-tier models
            system_instruction=RULES,
            input=user_text,
        )
    st.markdown(result.output_text)

requirements.txt: streamlit and google-genai. Add pandas or pypdf if you need them. Add .streamlit/secrets.toml to .gitignore.

Run locally: run pip install -r requirements.txt, put GEMINI_API_KEY = "..." in .streamlit/secrets.toml, then run streamlit run app.py.

Deploy free: push to GitHub, sign in at share.streamlit.io with GitHub, and create an app from your repo. In Advanced settings, then Secrets, paste your key, then deploy. You'll get a public link for your CV.

Info: This snippet follows Google's official Gemini API quickstart (google-genai SDK) as of September 2026. If it errors, compare it with the current quickstart, or paste the error into Cursor and ask it to explain.

Project 01: Bursary Finder

What it does: The student enters their degree, university and field of study. The tool shows bursaries they may be eligible for, with plain-English reasons, closing dates and official links.

Who it helps: Matrics, first-years, TVET and university students. Difficulty: Beginner. Version 1: one weekend.

The smart way to build it: don't let the AI "find bursaries" from memory, because it can invent them. Do it in three steps instead:

1. You collect real bursaries into a table from official sites.
2. Your code filters the table.
3. The AI only explains the matches it gets.

This is beginner retrieval-augmented generation (RAG). Use that word in your interview.

• No-code: Build the table in Google Sheets. Paste the instructions below plus the table into AI Studio, and test with fake student profiles.
• Code: Save the table as bursaries.csv. Use pandas to filter, for example df[df["fields"].str.contains(field, case=False)]. Add a Streamlit form and Gemini to explain the matches.
• Table columns: name, provider, fields, study_level, requirements, closing_date, official_link, last_checked.

System prompt:

You help South African students find bursaries. You will receive (1) a student profile and (2) a table of bursaries.
- ONLY use bursaries from the table. Never invent bursaries, amounts, requirements or dates.
- For each possible match: say why it MAY fit, which requirements the student must still check, the closing date and the official link.
- If a closing date has passed, clearly mark it CLOSED.
- If nothing matches, say so and suggest checking NSFAS and their institution's financial aid office.
- End with: "Always confirm details on the official website — bursary info changes often."

Roadmap:

• Version 1 (weekend): On Saturday, collect 15–20 real bursaries and build the form and filter. On Sunday, add the AI explanation, test 5 fake profiles, deploy, and write your README.
• Version 2 and 3: Auto-mark bursaries CLOSED and show "days left". Show "last checked" dates, add a document checklist per bursary, and offer explanations in more South African languages (checked by first-language speakers). Add a "report outdated info" button.

Data sources (official):

• nsfas.org.za: government student funding for public universities and TVET colleges.
• careerhelp.org.za/funding: DHET Career Development Services bursary listings with closing dates. Career helpline: 086 999 0123.
• funzalushaka.doe.gov.za: Funza Lushaka bursary for student teachers.
• mqa.org.za/bursaries: Mining Qualifications Authority. Also check the SETA for your field. dhet.gov.za lists the SETAs.
• Your institution's financial aid office, and company bursaries on the company's own website.

Warning: Responsible AI. Link every result to an official source and show when you last checked it. Warn users that the info may be outdated and they should confirm it on the official site. Real example: when I checked careerhelp.org.za on 30 September 2026, several listed bursaries closed that same day. Say "may be eligible", never "you qualify". Don't collect ID numbers or income details (think POPIA). Warn users about any "bursary" that asks for a fee over WhatsApp or SMS. Check a site's terms before you scrape it. For Version 1, collect by hand.

Show it off: In the README, include the problem, a screenshot, how the data is collected, your "only official sources" rule and known limitations. For the demo, run a profile like "BEng Mining, 2nd year" through the tool. In interviews, explain why you didn't let the AI answer from memory.

Project 02: AI Scam Detector

What it does: You paste a suspicious job offer, bursary message, email or WhatsApp. The tool flags common scam warning signs, explains why each one is a red flag, and says what to verify before you trust it.

Who it helps: Job-seekers, students and family. Difficulty: Intermediate. Version 1: one weekend.

Why it matters: The Department of Employment and Labour has warned about fake job offers that ask for "refundable fees". It says that under the Employment Services Act, no one may charge a work seeker a fee for employment services.

How it works (rules plus AI): Simple code rules catch obvious phrases. Then the AI reads the whole message, quotes the suspicious parts and explains them. Every result ends with what to verify and where.

RED_FLAGS = {
  "asks for money": ["admin fee", "refundable", "deposit"],
  "asks for secret codes": ["otp", "pin", "password"],
  "pressure / urgency": ["urgent", "within 24 hours"],
}
text = msg.lower()
found = [label for label, words in RED_FLAGS.items() if any(w in text for w in words)]

Warning signs to teach it:

• Asks you to pay a fee for a job, interview, training or bursary.
• Asks for your OTP, PIN or password.
• Uses urgency or pressure.
• Too good to be true (no interview, high pay).
• Free email address or a WhatsApp-only "HR".
• Asks for your ID or bank details up front.
• Spelling mistakes or shortened links.
• A real organisation's name with a different phone number.

No-code: Put the prompt below in AI Studio. Test it on 10 scam-style and 10 legit messages you write yourself, and record where it gets things wrong.

Code: Starter skeleton plus the rule check. Show the rule hits first, then the AI explanation.

System prompt:

You help South African students spot WARNING SIGNS in job offers, bursary messages, emails and WhatsApp messages.
You can NEVER be certain whether a message is safe or a scam. Never say "this is 100% safe" or "this is definitely a scam".
Reply with: 1) Warning level: Few signs / Some signs / Many signs. 2) Each warning sign: quote the exact words and explain why.
3) What to verify, using OFFICIAL channels (the organisation's official website or number that the student finds themselves, not contacts in the message).
4) If money, an OTP or banking details were already shared: contact your bank's official fraud line immediately and report it to SAPS.

Official channels your tool should point to:

Situation	Advice
Job offer asks for a fee	Don't pay. Verify the vacancy on the employer's official website. For anything that uses the Department's name, check labour.gov.za.
"Bank" asks for your OTP or PIN	Don't share it. Call the fraud number on your card or in your banking app. SABRIC advises: pause, verify through official channels, and protect your PINs, passwords and OTPs.
Money already shared	Call your bank's official fraud line immediately, then open a SAPS case. For anonymous tip-offs, use SAPS Crime Stop: 08600 10111.
Identity may be stolen	Contact the Southern African Fraud Prevention Service: safps.org.za, 011 867 2234.
Bursary or NSFAS message	Check only on the official site (for example nsfas.org.za) or with your financial aid office. Never use the links in the message.

Roadmap:

• Version 1: Write a test set of 15 scam-style and 15 legit messages, all fake examples. Build the text box and rules, add the AI, count false alarms and misses, then deploy.
• Version 2 and 3: Structured JSON output for neat warning cards, screenshot upload, a link checker, and South African languages (tested with first-language speakers). Add an accuracy table per version to your README.

Warning: Responsible AI. Never claim certainty. Show warning signs, not verdicts, because a clean result doesn't mean the message is safe. Keep a permanent disclaimer on screen. Tell users to remove names, ID numbers, account numbers and phone numbers before pasting, because free API tiers may use inputs to improve the provider's products. Don't store users' messages. Double-check every phone number and link you put in the app.

Show it off: Put a test-results table in the README (for example "V1: 24/30, then V2: 28/30"; use your real numbers). For the demo, paste a fake "R250 admin fee" job offer. In interviews, talk about false positives versus false negatives.

Project 03: AI Study Assistant

What it does: You upload your own notes. You can ask questions, get concepts explained simply, and generate quizzes, all based only on what's in your notes.

Who it helps: Every student at exam time. Difficulty: Intermediate. Version 1: one weekend.

How it works:

1. Upload a PDF or text file.
2. Extract the text with pypdf.
3. Send the notes and the question to Gemini with strict "only use these notes" rules.
4. Quiz mode generates 5 multiple-choice questions.

from pypdf import PdfReader

file = st.file_uploader("Upload your notes (PDF)", type="pdf")
if file:
    notes = "".join(p.extract_text() or "" for p in PdfReader(file).pages)
    question = st.text_input("Ask a question about your notes")
    # send: "NOTES:\n" + notes + "\n\nQUESTION:\n" + question

Scanned or handwritten notes have no extractable text, so start with typed notes.

No-code: Try Gemini Notebook (formerly NotebookLM). Upload your notes as sources, then ask questions and generate quizzes and flashcards. The free tier has daily limits. Use it as a benchmark, then build your own version with a twist (your module, your language, exam-style questions).

Code: Starter skeleton plus pypdf and two buttons, Ask and Quiz me.

System prompt:

You are a friendly study buddy for a South African university or TVET student.
- Answer ONLY using the NOTES provided. If the answer is not in the notes, say "That's not in your notes" and suggest what to look up.
- Explain simply first, then give one short example. Mention which part of the notes you used.
- When asked for a quiz: create 5 multiple-choice questions (A-D) based only on the notes, then list the answers with a one-line explanation each.
- Help the student learn. Do not write assignments or assessed work for them.

Roadmap:

• Version 1: Use one module's typed notes (5–20 pages). Build upload and text extraction, add "Ask", and test 10 questions, including 3 that aren't in the notes. Then add "Quiz me" and deploy.
• Version 2 and 3: For big notes, add chunking plus retrieval (proper RAG; look up "embeddings"). Add scored interactive quizzes, weak-topic tracking, a "first-year vs exam mode" toggle, and home-language explanations.

Warning: Responsible AI. Only upload material you're allowed to use. Your own notes are safest. Textbooks, paid guides and lecturer slides are often copyrighted, so keep them out of public repos, demos and apps. Check your institution's academic-integrity and AI-use rules, because they differ between institutions and even modules. AI can still be wrong, so cross-check important facts. Don't upload documents containing other people's personal info.

Show it off: In the README, show a screenshot of the assistant refusing an off-topic question. That's a feature. In interviews, explain the difference between a general chatbot and a grounded assistant.

Which project should you pick?

If you…	Start with
Are brand new to coding	01. Bursary Finder (mostly data and filtering)
Like safety, fintech or cybersecurity	02. Scam Detector (testing, rules and AI)
Want something you'll use daily	03. Study Assistant (you're user number 1)

More beginner project ideas

Inspired by the Mpumalanga AI Student Hackathon

The student teams there built real examples: SignSure (identity verification for digital signatures), AgricShield AI (helping agriculture recover from market disruptions), and MineLink & MineGuard (mining safety). Their full builds are ambitious, but you can start with a small spin-off.

1. "Before You Sign" explainer (SignSure-inspired). Paste a lease, contract or bursary agreement and get a plain-English summary, key obligations and questions to ask. It helps students signing their first lease or bursary contract. Tools: the starter skeleton and Gemini. It must say "this is not legal advice" and must never ask for ID numbers.
2. Crop and market helper (AgricShield AI-inspired). A small-scale farmer describes a crop problem or a market question. The tool gives possible causes and questions to take to an agricultural extension officer. It helps farming families and agri students. Tools: an AI Studio prompt, then Streamlit. It must not give definitive pesticide or dosage advice.
3. Mine safety quiz coach (MineLink and MineGuard-inspired). Turn public safety procedures or your own course notes into checklists and quizzes. It helps mining and engineering students. Tools: the Study Assistant code. It supplements official safety training and never replaces it.

More ideas

4. Power and water outage helper. National load-shedding has been suspended since May 2025, according to Eskom (497 days as of late September 2026), but Eskom still runs localised load reduction in parts of Gauteng and KZN, and municipalities announce water outages. You enter your area, and the tool turns the official schedule you've entered into a plain summary and reminders. It helps students studying at night. Tools: a CSV you update from Eskom's load reduction page or municipal notices, plus Streamlit and Gemini.
5. CV and cover-letter reviewer. Paste your CV (remove your ID number, address and phone first) plus a job advert. You get specific feedback and a tailored cover-letter draft. It helps final-years and job-seekers. Tools: the starter skeleton. The prompt must never invent experience or qualifications.
6. Local-language explainer. It explains lecture concepts or official letters in isiZulu, Sepedi, isiXhosa, Afrikaans and other languages, keeping technical terms in English. It helps first-generation students and families. Tools: Gemini with a clear prompt. AI quality varies across South African languages, so test with first-language speakers.
7. Township small-business assistant. A spaza, salon or food seller keeps a price and stock list in Google Sheets. The AI drafts WhatsApp specials, simple profit summaries and customer replies. It helps local entrepreneurs, maybe your own family. Tools: Google Sheets and AI Studio (no-code), then Streamlit.
8. Taxi and public-transport route helper. Build a community-collected table of routes, ranks, hand signs and approximate fares for your town. The user asks "How do I get from X to Y?" It helps new students in unfamiliar cities. Tools: CSV, pandas and Gemini. Label fares and routes as community estimates with a "last checked" date.
9. Mental-health resource finder. It points students to the right support: campus counselling, the Higher Health 24-hour crisis helpline (0800 36 36 36), and SADAG. It helps any student who is struggling. Tools: a curated list and a simple Streamlit app. It is not a therapist. It must never diagnose, and if someone mentions being in danger it must show crisis numbers immediately. Keep the AI part minimal.

GitHub basics and a README template

Browser-only way:

1. Create a free account at github.com with a professional username.
2. Click New repository, give it a clear name (for example ai-scam-detector) and tick "Add a README".
3. Choose Add file, then Upload files.
4. Write a clear commit message, for example "Add V1 rule-based checks".
5. Pin the repo on your profile.

New to it? Start with GitHub's Hello World guide, or use GitHub Desktop (free).

Command-line way:

git init
git add .
git commit -m "First version of bursary finder"
git branch -M main
git remote add origin https://github.com/YOU/REPO.git
git push -u origin main

Warning: Before every push, check that .streamlit/secrets.toml is in .gitignore. Commit little and often: 20 small commits tell a better story than one giant upload.

README template:

# [Project name]
One line: what it does and who it's for.
Live demo: [link] · Demo video: [link] · ![screenshot](screenshot.png)

## The problem          – why it matters in South Africa
## What it does         – 3–5 bullet points
## How it works         – tools used + simple steps/diagram
## How to run it        – install → add API key to secrets → streamlit run app.py
## Testing              – what you tested + results
## What went wrong      – honest bugs and fixes
## What I learned       – technical + non-technical
## Limitations & responsible use
## Next steps
## About me             – name, course, institution, LinkedIn

The What went wrong and What I learned sections matter most. Anyone can copy code, but your thinking is what gets you hired.

Show it off: interviews, LinkedIn and a demo video

Interview structure (60–90 seconds): Problem, then what you built, then what went wrong, then what you learned, then what's next.

Example answer. Swap in your real details and numbers. "Last semester I noticed classmates kept getting fake job and bursary messages on WhatsApp, some asking for refundable admin fees. So I built an AI Scam Detector: you paste a message and it flags warning signs, explains why each one is risky, and tells you what to verify through official channels. I built Version 1 in a weekend using Python, Streamlit and the Gemini API, and it's live with the code on my GitHub. What went wrong: my first version was too confident and called some real job adverts scams. I made a test set of 30 messages, found the false alarms, and redesigned it to show warning signs instead of a yes/no verdict, plus I added simple rule checks. I learned that testing and honesty about limitations matter as much as the AI itself. Next I want to add isiZulu support and screenshot uploads."

Prepare for:

• Why this problem?
• How did you test it?
• What happens when the AI is wrong?
• How did you protect privacy and your API key?
• What would you do with more time?

LinkedIn post template:

I built my first AI project: [name]
The problem: [who struggles, and why]
What it does: [feature] [feature] [feature]
What went wrong: [honest lesson]
What I learned: [1–2 takeaways]
Built with Python, Streamlit + Gemini API.
Try it: [link] | Code: [GitHub link]
Inspired by @africantechbro.ai #AI #SouthAfrica #StudentProjects

60–90 second demo video:

• 0–10s: the problem, shown with a fake example.
• 10–45s: a screen recording of the app doing its main job.
• 45–65s: one thing that went wrong and how you fixed it.
• 65–90s: what's next, plus "link in bio or comments".

Your 4-week plan

About 4–6 hours a week.

Week 1: Set up and choose

• Create GitHub and Google AI Studio accounts and get your API key
• Apply for the GitHub Student Developer Pack
• Pick one project and write the problem in one sentence
• Talk to 3 students who have that problem
• Try the no-code version and save your best prompt

Week 2: Build Version 1

• Get the starter skeleton running (laptop or Colab)
• Add the core feature and system prompt
• Create a test set (10–30 examples)
• Push to GitHub and commit every session
• Keep a "what went wrong" note

Week 3: Test, fix, deploy

• Deploy on Streamlit Community Cloud
• Watch 5 real users try it
• Fix the top 3 problems they hit
• Re-run your test set and record the results
• Add one upgrade plus responsible-AI disclaimers

Week 4: Show it off

• Finish the README with screenshots
• Record a 60–90 second demo video
• Post on LinkedIn and tag @africantechbro.ai on TikTok or Instagram
• Add the project to your CV (live link and GitHub)
• Practise your interview answer out loud 3 times, then apply

Tip: Stuck? Read the error slowly, then paste it into Cursor or AI Studio and ask it to explain like you're a beginner, then fix it. If that doesn't work, make the app smaller until it runs. Write down what broke, because that's your interview story.

Now go build

Pick one project and build the small version this weekend. Put it on GitHub, and write down what broke and what you learned. That's already more than most CVs can show.

• Follow @africantechbro.ai on TikTok and Instagram.
• Tag me in your demo video. I want to see what you build.
• Share this guide with your class group.

Links and sources

Checked 30 September 2026.

Building

• github.com
• education.github.com/pack
• aistudio.google.com
• Gemini API pricing
• console.groq.com
• colab.research.google.com
• streamlit.io
• gradio.app
• huggingface.co/pricing
• code.visualstudio.com
• cursor.com/pricing
• notebooklm.google.com
• python.org

Bursaries

• nsfas.org.za
• careerhelp.org.za/funding
• funzalushaka.doe.gov.za
• mqa.org.za/bursaries
• dhet.gov.za

Scams and privacy

• labour.gov.za
• sabric.co.za
• safps.org.za
• SAPS Crime Stop: 08600 10111
• inforegulator.org.za (POPIA)

Wellbeing and utilities

• higherhealth.ac.za (0800 36 36 36)
• sadag.org
• Eskom load reduction

Info: Free tiers, limits, prices and product names change often. NotebookLM was renamed Gemini Notebook in July 2026. Always check official pages and current pricing. This guide is general information for learning, not financial, legal, medical or security advice. Confirm bursary details with the funder. Scam tips help you spot red flags but can't guarantee anything is safe. If you've lost money, contact your bank's official fraud line immediately and report it to SAPS.