100k chats on SensAI

In 2024, we noticed a glaring problem at HyperVerge Academy. We prepared many practice problems which our learners were diligently attempting. But, they received little to no feedback.

Why? Because it was not possible to review thousands of submissions by a small team of less than 10 teachers. Learners never got to know their mistakes and the space for asking doubts was limited.

A missed opportunity for learning and growing.

But, for the first time, personalized learning at scale was no longer just a wish, thanks to the advancements in AI. We built a hacky prototype for a learning platform, SensAI, where teachers can delegate assessments to AI and create their own personalized AI tutors to support their learners 24x7.

Now, 15 months later, AI is no longer just a promise for improving learning outcomes. It is already doing it, not by giving them the answers but by asking the right questions for them to think about and meeting them at their level. Learners no longer need to wait for hours to get unstuck or hesitate to ask what they perceive as “stupid questions”.

The image shows one such conversation where the AI tutor does not point out the mistake straight away but guides the learner on how to think about the problem in a step by step manner.

Our learners are selected after a rigorous interview process that checks for motivation, curiosity and hunger. Once the feedback loop was shorter, they started acting on it - attempting more than before and striving to complete their tasks. Even their doubts have reduced.

Teachers don’t need to guess which learners are struggling or grind through a mountain of submissions to identify their strengths and weaknesses. Thanks to the automated assessments, they know precisely where a learner is struggling which helps them focus their limited time on what can drive the most impact. Seeing the learners' engagement and with more of their time freed up, teachers are also motivated to be more creative with their assignments.

This is not limited to technical topics and extends to what we call “power skills” as well - critical thinking, reading, listening, writing, speaking - through the power of AI that not only reads but also listens.

Of course, it is not perfect and makes occasional mistakes. But that’s the thing. It does not need to be perfect (except in certain domains). Through rigorous evals and a strong human-in-the-loop design, we can identify mistakes before they appear for real users and even if they do, we can mitigate them through continuous improvements. These incremental improvements, compounded over time, can build a system which is highly accurate, reliable and trustworthy.

These are the kind of AI use cases that personally excite me. The ones that uplift our potential and enable us to do more of what we find meaningful.

Humans + AI, not AGI.