Project - Rebuilding an AI tutor for real scale
Luma Learn engaged Zebra Labs to rebuild a fast-moving AI tutoring product after the original MVP architecture started to buckle under real usage and growth.
- Client
- Luma Learn
- Year
- Service
- Fractional CTO and system rebuild
Overview
Luma Learn had already proved there was real demand. The product had grown quickly, usage had spiked hard, and the AI tutor was reaching students across multiple countries and channels.
The problem was not whether the product mattered. The problem was that the MVP architecture had been built for speed, not durability, and the system started to show that strain under real load.
What Zebra Labs delivered
Zebra Labs stepped in to redesign and rebuild the product foundation around a more durable operating model.
The work included:
- migrating the system away from the original single-provider bottlenecks
- introducing clearer service boundaries and async message processing
- improving operational visibility so basic performance and product questions were answerable without painful manual investigation
- rebuilding the system in a way that supported continued delivery, not just emergency stabilisation
The result was a more reliable platform with a stronger path for future growth, multi-country delivery, and ongoing product development.
Why it mattered
This is one of the clearest examples behind Zebra Labs' system rebuild positioning.
Luma Learn was not a greenfield build and it was not a hypothetical technical-debt story. It was a live product with traction, urgency, and a codebase that needed to be reworked before it became the thing slowing the business down.
That combination shaped how the rebuild was approached: protect continuity, reduce technical risk, and create a system the team could actually keep shipping on top of.
Technologies used
- AWS
- Python
- FastAPI
- PostgreSQL
- Redis
- Next.js
Luma Learn operates at serious scale across South Africa, and William built the architecture that makes that possible. Modular, clean, AWS-native, and designed with our actual product needs in mind, not just best practices in isolation. He's a technically exceptional engineer who genuinely invests in understanding what you're building and why.