AI Research Foundations · Post-workshop period

Capstone Project Hackathon

Work in pairs to build a machine learning solution to a real-world problem in your own context, following the Responsible Innovation Roadmap in Course 08.

See your deadlines Open Course 08

Overview

Build your AI solution for real-world impact

During the post-workshop period, you and a partner from your cohort build a machine learning solution to a real-world problem in your own context. You apply responsible innovation practices from problem definition through to implementation, following the roadmap you walked through during the workshop.

Who takes part

All AI Champions and TAs in each cohort.

Teams

Pairs within the same cohort.

Categories

Classification, summarisation, conversational assistant, a combination of two, or your own path.

Deliverables

A capstone project proposal, and a GitHub repository or Google Drive folder with your implementation.

Recognition

The top three projects in each cohort are recognised.

Support

Contextualised datasets, code recipes, and technical support on Discord.

Course 08

The Responsible Innovation Roadmap

Every pair follows the same eight steps, from the problem you want to solve to a full plan you can share. Course 08 on Google Skills walks you through each one.

Open Course 08 on Google Skills
The Responsible Innovation Roadmap: eight steps from defining your problem statement to reviewing your full plan

Your way into the Community of Practice

Completing all the available courses and your capstone project during the post-workshop period is a prerequisite for joining the Community of Practice.

Capstone Project Roadmap

Your capstone, step by step

  1. 1

    Form your pair

    Team up with another AI Champion or TA from your cohort and register your pair.

  2. 2

    Choose your idea

    Pick a project category and a real-world problem that matters in your context.

  3. 3

    Follow the Responsible Innovation Roadmap

    Work through the eight steps in Course 08, from your problem statement to your full plan.

  4. 4

    Submit your capstone project proposal

    A document covering your problem statement, data collection plan, impact statement, stakeholder analysis, governance blueprint, sustainability plan and a one-sentence elevator pitch.

    Deadline: Two weeks before your post-workshop period ends
  5. 5

    Build and submit your implementation

    A GitHub repository or Google Drive folder with your code, Colab notebooks, data card, model card and any other documentation. GitHub is highly recommended, but Google Drive is also accepted.

    Deadline: The end of your post-workshop period
  6. 6

    Judging and recognition

    The AIMS South Africa team judges the projects and selects the top three in each cohort. FATE Foundation organises the recognition.

  7. 7

    Join the Community of Practice

    Completing all the available courses and your capstone project unlocks your place in the Community of Practice.

The eight roadmap steps

1

Define your problem statement

The problem, who it affects and why it matters in your context.

2

Create your data card

Where your data comes from, what it contains and its limitations.

3

Develop your impact statement

The change your solution should make and how you will measure it.

4

Conduct a stakeholder analysis

Who is affected by or has influence over your solution.

5

Draft your governance blueprint

How risks, privacy and accountability will be managed.

6

Build your sustainability plan

How the solution can be maintained and grow after the hackathon.

7

Share your work

Present your plan and solution to others.

8

Review your full plan

Check that every part is complete and consistent before you submit.

Submit your proposal

Due no later than two weeks before your cohort's post-workshop period ends.

Link coming soon

Submit your implementation

Due by the end of your cohort's post-workshop period.

Link coming soon

Team formation

Find your partner

Work in pairs

Every capstone project is built by two people: AI Champions, TAs, or one of each.

Stay in your cohort

Both partners come from the same cohort. Your partner can be from your own university or another university in your cohort.

Study buddies too

Your partner is also your study buddy: keep each other on track to complete all the available courses before your deadline.

Tips for a strong pair

  • Choose a partner whose skills complement yours, for example one stronger in coding and one in the problem domain.
  • Agree early on when and how you will meet, and share the eight roadmap steps between you.
  • Pick a problem you both care about and can find or create data for.

Register your pair

Once you have a partner, register your pair so the team knows who is working together.

Link coming soon

Judging criteria

How projects are judged

The AIMS South Africa team judges submitted projects and selects the top three in each cohort, considering:

  1. 01

    Impact

    The potential impact of the proposed solution on its target users and their local context.

  2. 02

    Responsible innovation

    How well responsible innovation practices are applied to assess the risks, potential harms, fairness, privacy, stakeholders and sustainability of the solution.

  3. 03

    Dataset

    Uses a relevant, well-sourced dataset (existing, collected or synthetic) and documents it in a data card.

  4. 04

    Model

    Uses a relevant model or machine learning technique, with a model card covering its architecture, intended use, training data, evaluation results, limitations and risks.

  5. 05

    Clean coding practices

    Clear structure, comments and reproducible steps, whether in Google Colab or Python scripts.

Recognition

The top three projects in each cohort will be recognised by FATE Foundation, and outstanding projects will be spotlighted in the Community of Practice.

Deadlines

Submission deadlines by cohort

Proposals are due two weeks before your post-workshop period ends; implementations are due on its last day.

CohortPost-workshop periodCapstone proposalImplementation
🇳🇬 🇬🇭 Nigeria Cohorts 1 and 2, Ghana Cohort19 Oct to 15 Nov 20261 Nov 202615 Nov 2026
🇰🇪 🇿🇦 Kenya Cohort 1, Pretoria Cohort19 Oct to 29 Nov 202615 Nov 202629 Nov 2026
🇰🇪 Kenya Cohort 226 Oct to 6 Dec 202622 Nov 20266 Dec 2026
🇿🇦 Western Cape Cohort16 Nov to 27 Dec 202613 Dec 202627 Dec 2026
🇳🇬 Nigeria Cohorts 3 and 4To be confirmedTo be confirmedTo be confirmed

Project ideas

Choose your path

Classification

Develop a language model or machine learning model for classification.

Example A maternal health SMS urgency classifier for community health workers in rural Kenya 🇰🇪

Summarisation

Develop a language model or machine learning model for summarisation.

Example A news headline generator for Swahili news 🇰🇪

Conversational assistant

Develop a chatbot or assistant that generates free-text responses to users' questions.

Example A personalised tutor for WAEC exams with support for Nigerian and Ghanaian languages 🇳🇬 🇬🇭

Combine two paths

Bring two categories together in one solution.

Example A customer support bot for ABSA 🇿🇦 that classifies the intent and urgency of a request in Zulu, then sends a solution

Your own path

Build any other project that solves a real-world problem in your context, using a machine learning model and responsible innovation practices.

More ideas to spark yours

A personalised AI tutor for Kenyan students who speak Swahili or Sheng 🇰🇪A multilingual chatbot for community health workers in rural Nigeria 🇳🇬A news summariser in Twi 🇬🇭A social media misinformation detection tool for Sesotho and Setswana 🇿🇦

Start from a recipe

Each recipe is a complete, working project you can run in Google Colab, then adapt to your own problem, language and data.

Classification Build a Customer Support Intent Classifier Open recipe Classification Classify Healthcare SMS by Urgency Open recipe Summarisation Build a News Headline Generator Open recipe Conversational Build a Customer Support Response Generator Open recipe Conversational Build a Multilingual Health QA System Open recipe Conversational Build Your Own Personalised AI Tutor Open recipe Fine-tuning Build a Domain-Specific Agricultural Small Language Model Open recipe Data Build a Synthetic Dataset for an African Language Open recipe

Resources

You don't have to start from scratch

Course 08 on Google Skills

The capstone course: it walks you through every step of the Responsible Innovation Roadmap and how to develop each part of your proposal.

Open Course 08

Capstone code recipes

End-to-end recipes from the AIMS South Africa team. Each one runs in a single Google Colab notebook on a free GPU, with a completed run to compare against.

Open the recipes on GitHub

Capstone datasets

Ready-to-use, contextualised datasets in English and African languages, so you don't have to start from scratch.

Browse the capstone datasets

Technical support

A dedicated #capstone-project channel will be created on each cohort's Discord server for your capstone questions. Report technical problems in #report-an-issue, and ask about the course material in #help-desk.

How to create your dataset?

A practical, step-by-step guide for beginners

How to Create a Dataset for Your AI Project

Learn how to source, collect, generate, organise, validate, and document data for an AI project.

A practical, step-by-step guide for beginners

How to Create a Dataset for Your AI Project

Learn how to source, collect, generate, organise, validate, and document data for an AI project.

  1. 1. Define your problem

    Identify what your model needs to learn or predict, and what a single example in your dataset should look like.

  2. 2. Choose how to obtain your data

    Explore public datasets from Zindi, Kaggle, Hugging Face and other reputable sources, manual data collection, and synthetic data generation.

  3. 3. Define your data format

    Decide on the columns, input features, target labels, and file format, such as CSV or JSON.

  4. 4. Collect or create your dataset

    Gather examples, generate synthetic samples, and annotate them where necessary.

  5. 5. Clean and validate your data

    Check missing values, duplicates, incorrect labels, inconsistent formats, class imbalance, and data quality.

  6. 6. Split your dataset

    Create training, validation, and test sets, avoiding data leakage.

  7. 7. Create a data card

    Document the dataset's source, purpose, collection process, contents, intended uses, limitations, and ethical considerations.

  8. 8. Save and share your dataset

    Organise the files, explain how to use them, and include reproducible instructions.

FAQ

Frequently asked questions

Can I work on my own?

No. Capstone projects are built in pairs, with both people from the same cohort. Your partner also doubles as a study buddy for the remaining courses.

Can my partner be from another university?

Yes, as long as you are both in the same cohort.

When are our deadlines?

Your proposal is due no later than two weeks before your cohort's post-workshop period ends, and your implementation by the end of it. See Deadlines for your cohort's dates.

Which dataset can we use?

You can use any relevant, well-sourced dataset. This could be an existing public dataset from platforms such as Zindi, Kaggle, or Hugging Face; a dataset you collect manually; or a synthetically generated dataset.

Make sure you document the dataset in a data card, including its source, purpose, contents, and any relevant limitations or considerations.

Do we have to use GitHub?

GitHub is highly recommended, as version control is a useful skill, but submissions through Google Drive are also accepted.

Who judges the projects?

The AIMS South Africa team judges the submitted projects and selects the top three in each cohort, using the judging criteria on this page.

What happens when we finish?

Completing all the available courses and your capstone project during the post-workshop period is a prerequisite for joining the Community of Practice, where outstanding projects will also be spotlighted.

Where can we get help?

A dedicated #capstone-project channel will be created on each cohort's Discord server for your capstone questions. Report technical problems in #report-an-issue, and ask questions about the course material in #help-desk.

Anything else? Email ai-research-foundations@aims.ac.za.