evidoria

← Back to browse

Good practice Imported

Kwame for Science — AI Teaching Assistant for West African Science Students

Ghana · Accra · See the Ghana profile · See the Accra profile

Evidence: Observational / pre–post Top 56% 47/100 · Ask Evidence Copilot about this practice

SuaCode.ai's Kwame for Science answers West African students' Integrated-Science questions via a web app aligned to WASSCE. An 8-month deployment logged 750 users in 32 countries and ~1,500 questions, with 87.2% top-3 answer accuracy against expert review (n=109).

87.2 %
Top-3 answer accuracy against expert review (8-month deployment (n=109 expert-reviewed questions))
750 users across 32 countries
Users during initial 8-month deployment (initial 8-month deployment)
~1,500 questions
Questions handled during initial deployment (initial 8-month deployment)
3,717 enrollments across 35 African countries
Kwame 2.0 enrollments (15 months)

Details

Maturity
Scaling
Promoter
SuaCode.ai (Kwame AI)
Period
2023
Keywords
AI tutoring, WhatsApp/web chatbot, secondary science education, WASSCE exam prep

Context

SuaCode.ai built Kwame for Science, an AI tutoring tool for West African secondary students preparing for the WASSCE Integrated Science exam, accessible for free via WhatsApp and a web app.

Objectives

To help students get accurate, curated answers and related past-exam questions for science queries, rather than open-ended generative text, as validated in a peer-reviewed evaluation of educational AI reliability.

Activities

The system takes natural-language student questions and returns curated passages and related past WASSCE questions; over 8 months it handled roughly 1,500 questions from 750 users across 32 countries (15 in Africa); a follow-up version, Kwame 2.0, added human facilitators and peer review to catch and correct AI errors, and grew to 3,717 enrollments across 35 African countries over 15 months.

Results

Accuracy against expert review (n=109) was 87.2% for top-3 answers, with lower accuracy for single top answers; the developers report expanded reach with Kwame 2.0 but no pre/post learning-gain data.

Conclusions

Kwame for Science shows technically validated answer accuracy and rapid usage growth via free WhatsApp/web access, but no published evidence yet demonstrates actual learning-outcome improvement for students.

Implementation
Indicative cost
Low (< €50k) — Not itemised in public sources; likely low given free WhatsApp/web-based delivery and small developer team.
Time to results
Medium (1–3 years) — Initial deployment ran 8 months; Kwame 2.0 follow-up ran 15 months with expanded enrollment.
Staffing & skills
SuaCode.ai development team, human facilitators and peer reviewers (added in Kwame 2.0) to catch/correct AI errors

Conditions for success

  • Free access via widely used channels (WhatsApp, web app)
  • Curated-answer approach rather than open generative text, to reduce error risk
  • Human-in-the-loop review added in Kwame 2.0 to catch AI mistakes

Common failure modes

  • Lower accuracy for single top-answer responses versus top-3
  • No pre/post learning-gain measurement published, so real educational impact unproven

Where it fits

Governance type
private EdTech developer (SuaCode.ai), not a government programme
Scale
regional (West/pan-African reach)
Income level
low- and middle-income (West Africa)

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful