Expanding Access to Education with Small AI Tutoring

CodeLink engineered an offline-first AI tutoring PoC for an international development organization. Designed for students in underserved communities, the Android app combines on-device AI, lesson-based learning, bilingual support, and voice interaction without requiring continuous internet access.

97.3% Corpus Alignment

AI responses aligned with the reference content in PoC testing.

99.1% Language Detection

Accuracy across English and French in the tested context.

18.0 Tokens/Second

Reported total throughput using Gemma 4 E2B on the benchmark Android device.

150 Test Prompts

Evaluated across three content domains and two grade levels by independent reviewers.

2 Languages Supported

English and French for learning and conversational AI.

4–8GB RAM Device Tiers

Assessed for AI feasibility, with principal benchmarks conducted on an 8GB Android device.

Client information background
Anonymous client logo

About the client

An international development organization focused on addressing global development challenges and improving access to essential services, including education. The organization explores innovative technologies to help underserved communities overcome infrastructure and resource limitations.

The challenge

CodeLink engineered an offline-first small AI tutoring PoC for an international development organization. Designed for students in underserved communities, the Android app combines on-device AI, lesson-based learning, bilingual support, and voice interaction without requiring continuous internet access.

Limited Access to Learning and Connectivity
Students in remote communities often lack reliable internet, teachers and educational resources. Conventional cloud-based AI tutors require connectivity for every interaction, while downloading large AI models can incur significant data costs, creating barriers to learning and technology adoption.
Resource-Constrained and Shared Devices
Affordable Android smartphones have limited memory, storage, and processing power, making it challenging to run advanced AI models locally while maintaining responsive, high-quality tutoring. At the same time, students have diverse language, literacy, and accessibility needs, and many share devices with others.
Ensuring Accurate, Safe and Relevant AI Responses
General-purpose AI models can generate inaccurate, inappropriate or off-topic answers. Educational applications require responses grounded in reliable learning materials, with age-appropriate explanations, safety safeguards and effective content governance, even without cloud-based controls.

Why CodeLink?

With extensive experience in AI engineering and enterprise software development, CodeLink has delivered intelligent applications and scalable platforms for global clients. This expertise positioned the team to explore Small AI as a practical approach to accessible, reliable learning in resource-constrained environments.

The result

The PoC demonstrated the feasibility of offline tutoring on affordable Android devices. Combining optimized language models, on-device RAG and bilingual learning, it established a technical foundation for delivering personalized educational support in low-connectivity environments.

Offline AI Tutoring and Local Model Distribution
CodeLink integrated small language models with offline retrieval-augmented generation (RAG), enabling students to ask questions, explore lessons and generate quizzes using locally stored content. The wider delivery framework also introduces Coach Mode, a peer-to-peer approach designed to reduce repeated data costs for model distribution and updates.
Personalized AI on Affordable, Shared Devices
CodeLink benchmarked quantized language models for low-cost Android devices, with Gemma 4 E2B achieving 18.0 tokens/second on an 8GB device in testing. The solution combines efficient on-device AI with bilingual English and French interactions, achieving 99.1% language detection accuracy. Offline speech recognition, age-aware explanations, and local-first student profiles support personalized learning across diverse needs and shared devices, even without internet access.
Curriculum-Grounded AI With Safety Controls
An offline RAG pipeline grounds AI responses in locally stored educational materials, achieving 97.3% corpus alignment in PoC testing. The broader framework also defines content validation, safety guardrails and controlled updates to support accurate, responsible AI tutoring.

Engineering Excellence.
Built for Enterprise and Institutional Innovation.

Contact Us
background

CodeLink Newsletter

Stay up to date with the latest insights on software engineering and AI strategy from CodeLink.

CodeLink

We partner with enterprises and institutions to turn complex technology challenges into reliable digital systems. Through expert software engineering, AI implementation, and disciplined delivery, we help organizations modernize operations, strengthen platforms, and deliver measurable business outcomes.

Contact Us

(+84) 2839 333 143info@codelink.io
Book A Discovery Call
2026 © CodeLink Limited.
All right reserved. Privacy Statement |