Sandeep Penmetsa Uses AI & Astra DB to Scale Education at Physics Wallah

Transcript

Physics Wallah is basically a tech platform with a mission to democratize education by making it affordable and accessible to all parts of the country. So we have been operational from 2020 and we started with JEE NEET preparation, but in the last four years we have been expanding to almost 25 plus verticals right from K5 learning til skilling.

We have all the offerings that are needed for students. Year long courses. They start around 4,000 rupees, which is by far the lowest in the industry because of which we have so many students, especially from tier 3, type 4 and rural parts of the country where the quality of teaching is not accessible.

So with Physics Wallah, the way we are able to do that is we have these teachers who have helped hundreds of students to get into JEE and NEET. Colleges have these mega classrooms where they handle close to 100k students on a daily basis.

With the scale that we have, which is close to 2.5 million users on a daily basis and close to 7.5 million users on a monthly basis. So the content delivery is being done by teachers. But then to crack these competitive exams, you need much more other aspects, as in how can you make sure that the students education is personalized?

And that's where we are leveraging AI. So we have built Alakh AI which is an educational suite of applications that are helping students in their entire learning journey. We have one of our flagship product called AI Guru which is a 24x7 available personal tutor where a student can come, he can ask academic, non academic product or support related queries.

So the way we are able to give him a concise answer is we have incorporated a vector database which has millions of numerical questions along with their video solutions. We have our proprietary textbooks also ingested and we also have thousands of micro learning videos which are of 2 to 5 minute time frame where each of the concepts are explained in a concise way.

So Astra is the vector database that we have used. And so it has been helping us in making sure that we are able to give concise and precise answers to the students. We did explore PGvector, but then it's just in a preview mode.

And then we also looked at Chroma, which is a native vector database. But then Astra fits our bill because it's hybrid. You can store the metadata information along with the embeddings, and that definitely helped us in expediting our whole product journey.

Regards to students. We have conceptualized the Alakh suite, and we are building innovative solutions around that. Apart from that, we are also looking at leveraging generative AI to enhance our internal efficiencies. So we have built an auditing framework for our sales counselors where they can understand.

The calls that they have taken with the customers and the areas of improvement, and are they actually adhering to the guidelines that Physics Wallah has. And apart from it, we are also looking at different avenues and how each of the businesses that are running in Physics Wallah can incorporate AI into their roadmap.

Most of the success that we have found was around RAG, as in we have generated so much of data over the last four years, so can we make sure that we leverage it to actually give more grounded answers? And Astra DB is our preferred vector solution partner for that.