The mission of our collaborative initiative is to democratize access to AI, ensuring equitable learning opportunities for all children and educators, regardless of their socioeconomic status.

Evidence-based & technically sound

Equitable and affordable

Implementable

and able to improve

learning at scale

Our North Star:

We want to support AI tools for low-and-middle income countries (LMICs) to be:

"We shape the world’s best technologies to transform education systems and help those learning the least"

We believe AI has the potential to radically transform education systems and improve learning.

SCALABLE QUALITY ASSURANCE SYSTEMS

REAL-WORLD INNOVATION AND SCALING

We are doubling down on two areas of high need/demand

Develop the solution / product

Pedagogical foundations

Product and AI evaluation

Safety and ethics

▪ ToC’s alignment to evidence / science

of learning

▪ Content quality

▪ Child safety / do no harm

▪ Data security and governance

▪ AI model benchmarks

▪ Evals on outputs

▪ Tech stack

▪ UX design

Get it in the hands of users

(under the right conditions)

RTM / access

Product uptake

Implementation

▪ B2C / B2B / B2G

▪ Infrastructure & device access

▪ Product cost & affordability

▪ Fidelity of implementation

▪ Supervision of product use

▪ User registration & trial

Experience & perceptions

Engagement

Understand the user experience

▪ Intuitive usability

▪ Satisfaction (NPS)

▪ Trust

▪ Dosage / time on task

▪ Engagement quality / use as intended

▪ Drop off rate

Intermediate user changes

Impact

(on outcome of interest)

▪ Cognitive

▪ Affective

▪ Behavioural

▪ Learning gains & subject mastery

▪ Eduficiency (teaching efficiency & productivity)

Cost-effectiveness

Measure the impact

▪ Impact per dollar spent (fixed / variable)

Quality assurance is central to our digital public good offer

1

2

4

3

Develop the solution / product

Pedagogical foundations

Product and AI evaluation

Safety and ethics

▪ ToC’s alignment to evidence / science

of learning

▪ Content quality

▪ Child safety / do no harm

▪ Data security and governance

▪ AI model benchmarks

▪ Evals on outputs

▪ Tech stack

▪ UX design

Get it in the hands of users

(under the right conditions)

RTM / access

Product uptake

Implementation

▪ B2C / B2B / B2G

▪ Infrastructure & device access

▪ Product cost & affordability

▪ Fidelity of implementation

▪ Supervision of product use

▪ User registration & trial

Experience & perceptions

Engagement

Understand the user experience

▪ Intuitive usability

▪ Satisfaction (NPS)

▪ Trust

▪ Dosage / time on task

▪ Engagement quality / use as intended

▪ Drop off rate

Intermediate user changes

Impact

(on outcome of interest)

▪ Cognitive

▪ Affective

▪ Behavioural

▪ Learning gains & subject mastery

▪ Eduficiency (teaching efficiency & productivity)

Cost-effectiveness

Measure the impact

▪ Impact per dollar spent (fixed / variable)

We support with content curation, benchmarks and evals to do controlled environment testing of AI tools

1

2

4

3

Developed three benchmarks: Pedagogy; SEND; visual reasoning.

Plus specific task evals: Gender balance; phonics alignment; EMIS data retrieval.

Plus agentic model for tracking products, including data privacy and security.

Creating a digital public good library of high-quality content for training/fine-tune/RAG: includes scalable evals for quality of textbooks; lesson-plans; story-books.

​

We’re developing benchmarks for educational use cases and filling key gaps—such as for Pedagogical knowledge

Performance of LLMs on pedagogical knowledge vs. cost

See our website for a complete / up to date list of LLMs assessed

Average human teacher score in Chile (50%)

KEY BENCHMARK NEEDS FOR AI-EDTECH

OUR PEDAGOGICAL KNOWLEDGE BENCHMARK

We’re identifying gaps for education which the AI community overlooks

Grade 7 Zambian visual reasoning exams

This matters, as early grade maths is learned through pictures.

KEY BENCHMARK NEEDS FOR AI-EDTECH

THE AI WORK IS CELEBRATING AI MODELS BEATING TOP MATH PROBLEMS—WHILE WE FIND THEY STRUGGLE WITH GRADE 7 ARICA VISUAL EXAMS

Develop the solution / product

Pedagogical foundations

Product and AI evaluation

Safety and ethics

▪ ToC’s alignment to evidence / science

of learning

▪ Content quality

▪ Child safety / do no harm

▪ Data security and governance

▪ AI model benchmarks

▪ Evals on outputs

▪ Tech stack

▪ UX design

Get it in the hands of users

(under the right conditions)

RTM / access

Product uptake

Implementation

▪ B2C / B2B / B2G

▪ Infrastructure & device access

▪ Product cost & affordability

▪ Fidelity of implementation

▪ Supervision of product use

▪ User registration & trial

Experience & perceptions

Engagement

Understand the user experience

▪ Intuitive usability

▪ Satisfaction (NPS)

▪ Trust

▪ Dosage / time on task

▪ Engagement quality / use as intended

▪ Drop off rate

Intermediate user changes

Impact

(on outcome of interest)

▪ Cognitive

▪ Affective

▪ Behavioural

▪ Learning gains & subject mastery

▪ Eduficiency (teaching efficiency & productivity)

Cost-effectiveness

Measure the impact

▪ Impact per dollar spent (fixed / variable)

We support innovators with small grants where we think there’s big public upside

1

2

4

3

Our learning by doing grants support targeted R&D

​

Plus support on evidence – and moving to offer support on evals.

​

To date small grants (<100k) focussed on digital public good components.

​

These projects have collectively reached millions of children with their innovations.

​

Plus we are investing in improving the training corpus including Voice-AI (in Tanzania).

Develop the solution / product

Pedagogical foundations

Product and AI evaluation

Safety and ethics

▪ ToC’s alignment to evidence / science

of learning

▪ Content quality

▪ Child safety / do no harm

▪ Data security and governance

▪ AI model benchmarks

▪ Evals on outputs

▪ Tech stack

▪ UX design

Get it in the hands of users

(under the right conditions)

RTM / access

Product uptake

Implementation

▪ B2C / B2B / B2G

▪ Infrastructure & device access

▪ Product cost & affordability

▪ Fidelity of implementation

▪ Supervision of product use

▪ User registration & trial

Experience & perceptions

Engagement

Understand the user experience

▪ Intuitive usability

▪ Satisfaction (NPS)

▪ Trust

▪ Dosage / time on task

▪ Engagement quality / use as intended

▪ Drop off rate

Intermediate user changes

Impact

(on outcome of interest)

▪ Cognitive

▪ Affective

▪ Behavioural

▪ Learning gains & subject mastery

▪ Eduficiency (teaching efficiency & productivity)

Cost-effectiveness

Measure the impact

▪ Impact per dollar spent (fixed / variable)

We are reimagining rapid efficacy studies to improve evidence at scale

​

1

2

4

3

We will manage a fund for rapid studies of real-world impact – initial project in SL. – expanding to full fund.

​

Have curated a set of shared, measurable indicators that can scale at lower costs, including learning outcomes.

We create evidence on what works

We curate what is out there

Track emerging evidence on AI tools for education.

​

Developed multi-agent systems to track products and find the public evidence claims.

​

(Beta) scaling AI-quality judgements against the ESSA frameworks

​

Developing web-app for Ministries/funders to check quality claims.

We are designing two district level projects to explore what is needed to take ideas into practice

Two countries, one county/ district each ⟶ reaching ~800 schools, 500,000+ learners impacted.

(Kenya: 281,973, Sierra Leone: 254,639)

WITH A FEATURE PHONE

A voice AI tutor

• She can call from a 2G phone – or it calls her when she misses class
• Discuss concepts she’s been struggling with
• Option to speak to her in her mother tongue
• Talk her through the lesson she missed

We will test how can AI support teaching and learning in a systematic way

A STUDENT

SMS practice exercises

• Personalised learning via SMS to reinforce subject learning

WITH A SMART PHONE OR TABLET

Learning platforms

• High-quality content available offline-first
• When 3G data is available, it updates the learner profile

​

A TEACHER

A QUALITY ASSURANCE OFFICER

Automatic marking

• Teacher takes a photo of students work, which connects to their EMIS learner profile
• A special ‘Smartpaper’ automatically marks student exercises & gives personalised feedback

WITH A SMART PHONE OR TABLET

Coaching & feedback

• Chatbot support for teaching practices
• Recording of lessons provides feedback on his instruction and classroom management practices

Lesson planning

• Plans his lessons for the day based on structured pedagogy
• Creates personalised lessons for different groups of students

Tailored guidance for each school

• Her tablet automatically pulls the EMIS data on the school she’s visiting, and AI combines it with policy guidelines to generate tailored advice for each school

Legend:

Underlying tech is not strong yet, but will be soon

Requires fully integrated solutions between school / home / EMIS

WITH A SMART PHONE OR TABLET

Tailored support to teachers

• She can see each teacher’s development areas and feedback from their AI coach, as well as student learning deficits at that school, and can coach teachers and head teachers on the areas they need most help with

School prioritization

• A dashboard on her tablet tells her which schools need most help, and helps her prioritize which ones to focus on

AI Trainings

Training Space

GenAI NGO Peer Learning Cohort

Africa ICT Right

Dignitas

EducAid Sierra Leone

Orkeeswa, Inc.

Phoenix Foundation

ReachAll

Rising Hope Girls Educational Foundation

The Agahozo-Shalom Youth Village

Yiya Engineering Solutions

Zimbabwe Information and Technology Empowerment Trust

​

​

team4tech.org | 14