AI in Education: How Zindua School is Transforming Learning with AI

AI is in a hype cycle – just like blockchain was 7 years ago. But hype doesn’t mean there’s no value. At Zindua School, we’re cutting through the noise, integrating AI where it truly enhances learning while being mindful of its limitations. We need to separate genuine potential from inflated promises.

For context: we’re a coding school that’s spent the last 5 years building our reputation on two key pillars. First, we deliver extreme personalisation in our Software and Data programs. Second, we’ve pioneered aggressive learn-now-pay-later financing options that have opened doors for talented individuals who might otherwise miss out. These approaches have consistently delivered strong graduate outcomes and made our programs genuinely accessible.

Now, as we build our Learning Management System 3.0, we’re taking a carefully considered approach to AI integration. We’re actively championing AI’s potential while being realistic about its current limitations. Our strategy involves gradually exploring AI advancements and finding creative ways to work around constraints to deliver real value to our students.

In this post, I’ll walk you through the three core areas where we’re integrating AI across our coding programs this year as we scale beyond Kenya. These aren’t just ideas – they are concrete initiatives we’re actively building and shipping. Let’s dive into how we’re making this happen. Note: Most of these features will be shipped in the last half of 2025 and early 2026.

1) Personalised Learning: AI Teaching Assistants for Quick Code Support

Extreme personalisation has always been Zindua’s biggest competitive advantage in the local market. It’s not just talk – we deliberately maintain full-time classes of 15 students and part-time classes of 7 students, and we’ve built our entire growth strategy around these numbers. Why? Because we’ve consistently seen that closer interaction with technical mentors leads to higher quality graduates and better job outcomes. This approach has given us incredible retention rates – we are talking 95% of students continuing from module to module.

But we’re not stopping there. Even with our low student-to-mentor ratio, we see an opportunity to push personalisation even further through AI teaching assistants. AI Teaching Assistants won’t replace our mentors—they’ll amplify them. By handling immediate support needs, they allow our technical mentors to focus on what truly matters: deep technical coaching and personalised guidance. Here’s exactly what our AI Teaching Assistants will do:

1. Provide real-time chat support when students hit roadblocks. Whether someone missed a concept in class or needs clarification on a topic, the AI will step in with explanations, direct them to relevant learning materials, and break down concepts based on their specific confusion points.
2. Offer strategic project support. We’re building these AIs to identify code issues and provide guided assistance – but here’s the key difference: instead of just spitting out solutions, they’ll offer step-by-step guidance that helps students think through problems. We’re in education, after all – our goal is to build problem-solving skills, not create solution-copying habits.
3. Bonus benefit: these AI assistants will help us improve retention in our shorter programs like free foundational courses and short courses. These programs don’t include personal mentors, but with AI support, we can offer a more supported learning experience without changing our core program structure.

To make this seamless, we’re integrating these assistants right where students need them – they’ll pop up directly in our learning management system on lesson and project pages, plus they’ll be accessible through Slack, our main communication hub. We’ve taken inspiration from Khan Academy’s Khamingo here, but we’re adapting the approach for our specific coding education context.

2) Project-based Learning: Automated Feedback for Daily Challenges and Weekly Projects

Let’s talk about how we’re revolutionising project feedback at Zindua. Here’s the reality: extreme project-based learning is the backbone of how we build job-ready graduates. Our current structure is intentionally intensive – students tackle daily challenges after each lesson (Monday through Thursday), complete weekly projects that synthesise each week’s content (Friday to Sunday), and build capstone projects at the end of each module, with a final capstone wrapping up the entire program. This heavy focus on practical work has proven effective, but it also creates a significant feedback burden on our technical mentors, even with our small class sizes.

Now, for some projects, there’s an easy solution. When we’re dealing with JavaScript and Python programming challenges that have clear expected outputs, auto-grading through Github Classroom is straightforward. Similarly, for machine learning projects, we can take a Kaggle-style approach – define clear test cases and evaluation metrics, and automate the assessment process.

But here’s where it gets interesting – and where AI comes in. The real challenge lies in evaluating projects that don’t have clear-cut right or wrong answers: frontend designs, full-stack applications, and data analysis dashboards. You can’t just run a test case to determine if a user interface is intuitive or if a dashboard effectively communicates insights. This is where we’re building AI agents to provide detailed, nuanced feedback that goes beyond what traditional autograders can offer. Our target implementation looks like this:

• Daily challenges get immediate AI feedback, in scenarios where auto-graders cannot be applied
• Weekly & capstone projects receive initial AI assessments, followed by human expert review within 3 days

This approach gives students quick initial guidance while ensuring they still get the depth of expert human review for their more substantial work. The AI isn’t replacing our mentors – it’s helping them focus their time on providing higher-level insights and guidance.

3) Quality Content: Adaptive Projects (Focus) and Personalised Content Remodels

Here’s something we’re really proud of at Zindua: we’ve invested significant time and resources into building market-aligned curriculums and content (always an ongoing initiative). Right now, our content mix includes text-only lessons, video-only lessons, and our ideal combo of text-and-video lessons. But with AI, we can push this even further.

We’re looking at leveraging AI to create truly personalised learning materials. Think about it – not every student learns the same way. We can use AI to generate different versions of our content: detailed cheat sheets for quick reference, condensed summaries for revision, and content tailored to different learning styles. Yes, AI video generation is still a hard nut to crack, but we’re starting with what’s achievable now – text and graphics generation – while keeping an eye on future possibilities.

Here’s where it gets exciting: we’re taking inspiration from GMAT’s FOCUS exam to build something game-changing – adaptive coding challenges. Instead of having all students work through identical challenges (except for capstones, which are always unique), we’re developing AI agents that can generate challenges based on individual student progress and capability. Here’s how it works:

• Students who grasp concepts quickly get more complex challenges to stretch their abilities
• Those who need more time get foundational challenges to build confidence and skills
• As students progress, the challenge difficulty adjusts until they reach our standard level

This isn’t just about making things easier or harder – it’s about creating a learning path that adapts to each student’s pace while ensuring everyone reaches the competency levels needed for the job market. What we’re building isn’t just another AI-powered learning platform – it’s a system that understands and responds to individual learning journeys. We’re starting with these core features because they deliver immediate value while setting us up for future innovations as AI technology evolves.

Looking Ahead: Beyond Our Core Focus Areas

While we’re laser-focused on delivering our three core AI initiatives this year, we’re already seeing promising applications in other areas of our operations. Take admissions screening, for instance. Our interview process is crucial – it’s how we ensure students land in programs that match their interests and background. AI could help streamline this matching process.

Then there’s career support. Before job placement, our graduates could benefit from tech-career-focused AI assistants helping them build their professional profiles – think LinkedIn optimisation, resume feedback, and even interview prep. We’re seeing lots of existing tools in this space that we could potentially build upon.

But here’s the thing – we’re intentionally keeping these as future considerations while we nail our core AI integration. At Zindua, we believe in doing fewer things better rather than spreading ourselves too thin. This brings us to an important discussion about AI’s current limitations and how we’re working to address them…

Working Around AI Limitations

Let’s be real about AI’s current constraints. As we build these features, we’re actively working to navigate around three key limitations. We’re learning from other players in the space and, frankly, doing a lot of creative problem-solving of our own.

Limitation 1: AI Hallucinations: A Real Challenge for Advanced Topics

In education, accuracy isn’t optional—it’s critical. While AI has improved, it still struggles with advanced topics like Data Structures & Algorithms, often making errors on less common problems. Since most AI models are trained on widely available code, they sometimes fail to capture nuanced problem-solving techniques—something we’re actively working to address.

Limitation 2: The Reasoning Gap

Yes, we’ve seen impressive improvements with models like Deepseek, ChatGPT o-models, and Gemini Flash in handling complex reasoning tasks. It’s way better than when AI would struggle with basic tasks like counting letters in words. But let’s be clear – we still need to be strategic about how we use AI for complex problem-solving. This means careful domain-based fine-tuning and being mindful of context window limitations, especially for problems that require numerous multiple steps.

Limitation 3: Bias and Data Limitations in Education

This is a big one for us. General internet data comes with its own biases and gaps, and these get baked into AI models. In education, where accuracy and fairness are crucial, we can’t just accept these limitations. We’re looking at expert model fine-tuning, using our internal content and expertise to reduce these biases. It’s not a quick fix, but it’s necessary for delivering the quality we expect at Zindua.

Conclusion

Here’s our bottom line: AI can absolutely transform how we deliver education at Zindua, but only if we’re smart about handling these limitations. As a data scientist who’s spent significant time in operations, I’m excited about tackling these challenges. We’ve got a solid development team working on our LMS 3.0, and we’re taking a gradual, thoughtful approach to AI integration.

As we work through these challenges and roll out new features, I’ll be sharing more about our specific solutions and learnings. Stay tuned – either we’ll figure out creative workarounds, or the competitive AI market will drive solutions that make these problems easier to handle. Either way, we’re committed to making this work for our students.