Skip to content
NewAI assistants

AI built by the people closest to the problem.

ACTS ML is a free, no-code AI builder from the African Centre for Technology Studies. Train a model on your spreadsheet, or build an assistant that answers from your own documents.

Start — it's free See how it works

No code and no card. Sign in with Google or email.

Predictive model

Predict or classify from a table.
AI assistant

Answers only from the documents you add.
Worked-example places · hover or tap a point to read its story
app.actsml.com/projects/create
The ACTS ML wizard on its first step, Define your problem, asking what you want to predict and whether a simpler, non-AI method could solve it

Two kinds of AI. One place to build them.

Most questions take one of two shapes. Some need a number or a category worked out from your data. Others need an answer found in your documents. ACTS ML handles both, without code.

Predict or classify from a spreadsheet

Predictive model

Bring a table: rainfall records, sales, survey results. The wizard helps you clean it, suggests an algorithm, then trains and tests the model with you.

  • Regression, classification and clustering
  • A recommended algorithm, with how confident it is and why
  • Results in plain terms: accuracy, F1 score, error in your own units
Training run summarysuccess
Run ID
TR-4C19A0E7
Algorithm
random_forest
accuracy
0.87
f1_score
0.84
Example run. Your numbers come from your data.

Answer questions from your own documents

AI assistantNew

Talk to the Assistant Builder and it sets things up with you: what the assistant is for, where its answers come from, and what it must never do. Then test it in chat.

  • Answers come only from the sources you add
  • You set what it must refuse and who it hands over to
  • PDF, Word, Excel, CSV, text files and public web pages
Assistant plan4 of 5 set
  • Purpose · Answer staff questions about the HR manual
  • Knowledge sources · hr-manual.pdf, leave-policy.docx
  • Refusal & escalation · Salary questions go to HR
  • Safety checks · No personal data in answers
  • Channels · Not set yet
Example plan, drawn from the Assistant Builder.

From a question to a working model, in six steps.

The wizard takes you through each step in order. As it says on the first screen: these questions aren't a test, they're here to guide you.

  1. Step 1, Define problem: Say what you want to know

    Write the question the way you'd say it. The wizard asks who it's for, what influences it, and whether a simpler, non-AI method could do the job.

    Define your problem step with questions about the prediction and its users

  2. Step 2, Choose data: Bring your data, or start from shared data

    Upload your own file, or pick an open dataset someone has shared on the platform. Every dataset comes with a Data Card describing where it came from.

    Explore Data page listing open-source datasets with Start project buttons

  3. Step 3, Prepare data: Clean it in the Data Studio

    Look through your table, apply the recommended fixes for missing or messy values, and commit each change. Every commit is a version you can go back to.

    Prepare your data step showing the Data Studio table with Preprocess, Features and Commit Changes

  4. Step 4, Algorithm: Get a recommendation, with reasons

    Choose the column you want to predict. ACTS ML recommends the type of model and a specific approach, says how confident it is, and explains why. You can change it.

    ML Algorithm

    Choose what to predict. ACTS ML suggests the rest.

    Target column

    rainfall_mm

    ML Algorithm Type (Recommended)

    RegressionHigh confidence

    The target is a number that changes smoothly, so a regression model fits.

    Random forest

    Recommended · handles missing values

    Linear regression

    Simpler · easier to explain

    Illustration of the ML Algorithm step.

  5. Step 5, Train model: Train, and keep every run

    Pick how to split your data (random, stratified or by time) and train. Each run gets its own ID, its metrics and a summary you can copy or export.

    Training run summary with run ID, algorithm, dataset version and metrics, plus Copy metrics and Export as CSV

  6. Step 6, Predictions: Test it with new values

    Type in a new case and get the prediction straight away. Retrain whenever new data comes in; older runs stay on record.

    Test model dialog with one field per input column, a Test button and an output panel

Messy data is normal. Cleaning it shouldn't be hard.

Real data has gaps, typos and columns in the wrong format. The Data Studio reads your table, suggests fixes column by column, and lets you apply one or all of them in a click.

Row Filters
Keep or drop rows by value, range or date.
Imputation
Fill gaps with the mean, median, mode or a value you choose.
Encoding
Turn categories and text into columns a model can use.
Scaling
Put numbers on the same footing: standard or min-max.
Create Features
Add, divide, bin, count words, or pull the month out of a date.
Commit Changes
Save every change as a version, with a note of why.
Data Studio · station_readings.csv

Recommended preprocessing

4 suggestions · review before you apply

ColumnSuggested fixAction
rainfall_mmFill with medianApply
stationOne-hot encodeApply
reading_dateExtract month, day of weekApply
temp_cStandard scalingApply
Initial commit → a91f2c0 4,812 rows · 11 columns
Illustration of Recommended preprocessing, with example data.
The View Data tab of a dataset in ACTS ML, with Preprocess, Features and Commit Changes controls above the table

Ask first. Build second.

Responsible AI isn't a separate checklist in ACTS ML. The questions sit inside the steps, at the moments where they change what you build.

  1. Before you build, define the problem with the people it affects

    The first step asks who the model is for, whether you worked out the question with them, and whether a simpler, non-AI method would do. Sometimes the honest answer is yes.

    Could a simpler, non-AI method solve this?

  2. When you add data, record how it was collected

    Uploading a dataset includes an ethics and consent step: was consent given, is it anonymised, is it representative, was it collected through someone the community trusts, in a language they understand?

    Was the data collected through a person the community trusts?

  3. When you build an assistant, keep answers to your own sources

    An assistant answers only from the documents and pages you add. You decide what it must refuse and where it should send people instead.

    Answers come only from sources you add.

A Data Card in ACTS ML showing collection details, geographic information, and data quality and compliance fields such as anonymised and informed consent
Every dataset has a Data Card: where it came from, who collected it, and whether consent was given.
Read the responsible-AI guidelines

Most AI is built far from the problems it's meant to solve.

A county officer in Kisumu knows which wards flood first. A tea factory manager in Kericho knows what a bad picking week looks like. That knowledge is already here. The part that usually has to come from somewhere else is the modelling.

ACTS ML puts the modelling in the same room as the knowledge. It's built in Nairobi by the African Centre for Technology Studies, which has worked on science, technology and innovation across Africa since 1988, so that the people who understand a problem can build the model for it themselves.

About ACTS
Nairobi's skyline seen across green trees on a clear morning
1°17′S 36°49′E · NairobiPhoto: Unsplash

Questions people are already asking.

Worked examples of what you could build, from the first question to the result.

All use cases
Storm clouds over dry grassland with scattered acacia trees1°48′S 37°38′E

Planting on a forecast, not a guess

Predictive model·Makueni County, Kenya

A ward agriculture officer wants to tell farmers how much rain to expect each week of the short rains.

Rolling green tea fields under a wide sky in Kericho0°22′S 35°17′E

Finding the plots that will fall short

Predictive model·Kericho County, Kenya

A tea co-operative has six seasons of delivery records and eleven extension officers to send out.

A woman working at a laptop and notebook by a window26°12′S 28°03′E

Finding the clause before the meeting

AI assistant·Johannesburg, South Africa

A compliance officer reads the same four hundred pages of by-laws every month, looking for one paragraph.

Questions, answered.

Something else? Ask the team or read the documentation.

Is ACTS ML really free?

Yes. You can sign up, upload data, train models and build assistants without paying, and you don't need a card to start.

Do I need to know how to code?

No. Everything happens through the wizard and the Assistant Builder. If you can work with a spreadsheet, you can use ACTS ML.

Who can see my data?

Your datasets are private unless you mark one as Open Source. Only then can other people find it in Explore Data and start projects from it.

What kind of data can I use?

For predictive models, tables such as CSV or Excel files. For assistants: PDF, Word, Excel, CSV, text and Markdown files of up to 25 MB each, and public web pages.

Can I use my model outside ACTS ML?

Not yet. Today you test models and make predictions inside ACTS ML, and test assistants in chat. Ways to share and deploy them are being built.

Who is behind ACTS ML?

The African Centre for Technology Studies (ACTS), a research centre based in Nairobi. ACTS ML is part of its work to make AI something people across Africa can build, not only use.

Start with one question.

Make a free account, bring your data, and find out whether a model helps your work.

Start — it's free Talk to the team