The world’s first AI teacher support tool

A WhatsApp trainer for teachers in the poorest schools.


Every teacher needs a friend

Every teacher needs a friend – to ask questions, get advice, and to discuss their day. For many teachers, working in schools that are miles from another, their support group is very limited. Yet this doesn’t have to be the case – whatsapp has made the world a smaller place, and allows people all over the world to communicate with each other at a very low cost.

The challenge then is how teachers can find each other – and get access to the latest knowledge in pedagogy, classroom management and general hints and tips to stay sane after a hard day in the classroom. For the poorest, they are lucky to get Continuous Professional Development training once per year – and then for a day or so with no opportunities to ask follow on questions.

This is where comes in – by combining cutting edge large language models with teaching expertise, the costs of getting information out to the poorest teachers can be dramatically lowered.

Sign up for our private BETA is in private BETA at the moment, let us know if you want to join our next round of users!

Simple, free to use with high quality results!


Ask a question


Receive an answer/links


Cut through the noise!

Fine-tuning large language models


To build this, Fab Inc. were granted early access to GPT-3 last year, and have been working to train and test a model to allow teachers to ask questions to a virtual coach or trainer to simulate communities of practise and teacher training.

The advantage of using GPT-3 is that it has been trained on a very large volume of data and new models are expanding over time – and when they come available the models can be retrained with the new, larger parameters, increasing the base and trained accuracy.

While these models are useful, they all require ‘fine-tuning’ where the products can be further trained by permitted users, who can then save that model and access in the future (which means you can invest in training and improving over time and it isn’t forgotten). This fine-tuning is done through showing the model a series of prompts and responses (in our case questions and answers), which Fab have curated through partnerships with UK and Sierra Leonean teachers and teacher trainers. This is then used with Reinforcement Learning Human Feedback (RLHF) where the teachers can rank the quality of the answers given by, to further improve the model.

Integrating pedagogy and real-world experience

What stage are we at?

This is currently in early testing stage, with Fab building the training data of teacher queries directly from teachers, with answers being curated by early grade specialists. Fab Inc. are working with Educaid, to get queries from teachers in Sierra Leone, alongside the initial questions from the UK literacy specialist. These questions are then answered by the literacy pedagogist and will be cross-checked by experienced teacher coaches, and used for training.

The initial testing consisted of around 200 questions asked from domains including knowledge, pedagogy teaching, and classroom management, with satisfactory automated answers[1]. 

While the team is at a very early stage the initial tests are positive. The team will work to collate more questions from teachers in Sierra Leone and enrich the model with these over time. As the model improves over time with more incoming data, so the more utilized it gets, the better it will work going forward but will require more training data and oversight in the first year or so.

Overall, this is an area that is advancing rapidly, but one in which the barriers to entry are falling rapidly, as the base levels of the models increase. However, the domain knowledge required to fine-tune models, and the time required to do this properly should not be underestimated, given the capacity for misinformation and misleading answers. That given, the potential to dramatically lower the costs of information provision to remote areas is high, and given the need to massively increase support for teachers around the world, one which can complement existing efforts very well.

[1] Simultaneously, the team is working on restricting the model to avoid any abusive responses by the chatbot which may otherwise appear given that the GPT-3 model is largely trained on internet data.


Simple to use and low-cost

The teacher’s friend is simple to use – an educator asks a question to which the chatbot provides an answer or links to more information so that the user can read more about the topic (such as a curated pdf or link to a webpage with more information). 

This will provide a low-cost way for educators to ask any education-related question, significantly reducing the noise that would have been obtained otherwise by internet browser searches which are influenced by advertising.

Work with us to keep open source and free to use! is open source and free to use, work with us to reach the teachers that can benefit the most!