Gianni Mazza Drives AI-Powered One-on-One Math Tutoring at Third Space Learning

Gianni Mazza, Technical Architect at Third Space Learning

Gianni Mazza
Technical Architect at Third Space Learning

Gianni Mazza is a seasoned Technical Architect with extensive experience in software engineering, cloud infrastructure, and platform modernization. Currently leading architectural initiatives at Third Space Learning, Gianni has played a pivotal role in refactoring a monolithic platform that serves thousands of teachers, students, and tutors across UK primary and secondary schools.

About the Company

Third Space Learning is making one-to-one maths support more accessible for schools. The maths attainment gap is growing; one-to-one support is one of the best evidenced interventions to help close this gap. But schools are increasingly required to do more with less, and it's not always possible for teachers to provide this level of support - especially not in way that works within their budget. Since 2013, Third Space Learning has worked with over 4,000 schools across the UK and US to provide cost-effective online maths tutoring to the pupils who need it most - developed by our team of former maths teachers and delivered one-to-one by STEM graduates and undergraduates with a passion for teaching. Our collections of evidence-based resources support schools everywhere to save time, build confidence and boost progress.

Nominate a Digital Champion today!

Transcript

My name is Gianni Mazza. I'm the technical architect at Third Space Learning. We have been in the market for more than 10 years offering quite a different solution. Other companies offer online solutions, but the most affordable solutions are groups teaching like tutoring for four or five people at the same time.
We are offering just one to one tuition math sessions and our tutors have always been in Asia instead of here in Europe. And that's what allows us to keep the price as low as possible. We are using AI in different few ways.
The first is trying to get our tutors better and better in what they are doing. And for this reason we built a platform to evaluate all the sessions. It's an AI math tutor and to bring the price even lower. Because the main problem we are having is trying to retain schools because as you can imagine, schools don't have a big budget for tuition. Our platform is for school and they pay us and they offer sessions to their children with data structs.
We are building a few things. The first is the memory for our AI tutor and that's the reason why I contacted those tracks in the first place. And we are planning to use the vector database for our AI evaluation platform to give our tutors insights in real time.
The vector database is really important for us for a few reasons. The first is the latency, the very low latency we can get data from the database. And the second is because we can contextualize our search in few milliseconds to give our model context to retrieve information. We considered a few other databases. So the very first attempt was using PGvector. It worked, but the latest wasn't great. So we tried other platform like MongoDB that has very recently introduced a vector database functionality. But data structure is definitely and Astra is definitely faster for us. We are using a few components on our AI tutor. For example, the main framework is LangChain and we are using LangChain to abstract the way we interact with the model. So we can use GPT 4.O, we can use Claude 3.5 Sonnet, we can use llama, we can use algorithm model like Phi-3 and without changing any code.