GCP – Inside Infinity Learn’s AI Tutor, powered by Google Cloud

Late-night study sessions, complex equations, and immense pressure: for millions of students in India, the IIT JEE Main exam and NEET for medical colleges are critical gateways for their futures. Exam success hinges on intricate reasoning within physics, chemistry, mathematics, zoology, and botany. These students require a clear, step-by-step understanding of the topics, not just the correct answers.

Recognizing this vital need for deep comprehension and problem-solving, Infinity Learn, a pioneer in tech-forward education, collaborated with Google Cloud Consulting to develop an AI tutor to assist students during exam prep. The goal wasn’t to merely create a question-answering tool, but a true tutor.

Acting as a custom search engine, the AI tutor allows students to learn through detailed guidance for solving math, physics, and chemistry problems. This searchable system aims to foster in-depth knowledge via Google Cloud’s Vertex AI Retrieval Augmented Generation (RAG) services. Using a cutting-edge Gemini 2.0 Flash model, the AI tutor enables students to independently find the answers they need across faculty resources, helping them truly grasp the how and the why behind solutions.

Powered by Agent Builder and Vertex AI RAG Engine

The biggest challenge early on in development was building a system that could do more than just anticipate exam questions. It needed to:

• Solve multi-step problems logically, mirroring a human tutor’s approach.

• Articulate each step in a way that promotes conceptual clarity.

• Serve a large user base (both free and paid) while balancing cost and accuracy.

• Analyze and self-correct errors to continuously refine performance.

In a collaborative effort, Google Cloud Consulting and Infinity Learn used Agent Builder for experimentation and RAG Engine on Vertex AI for refinement.

They began with Vertex AI’s Agent Builder, an out-of-the-box RAG system that allowed for rapid prototyping and quick iteration. Using Vertex AI Search within Agent Builder, they were able to supply the AI language models with Infinity Learn’s collective library of resources, ensuring answers were relevant, accurate, and grounded.

For further customization, they then implemented a do-it-yourself RAG engine with Vertex AI Search as retriever. This offered more flexibility when selecting the retrieval mechanisms and generation model, as well as granular control for fine-tuning the AI tutor’s accuracy and reasoning capabilities while using Google’s search infrastructure.

Beyond accuracy: Measuring real-world effectiveness

Following implementation, the team launched a thorough evaluation for the AI tutor. This model-based custom evaluation leveraged Vertex AI workbench and subject matter experts to confirm AI accuracy within a variation of up to 3-5% (as compared to human reviewer-based accuracy).

The evaluation process focused on measuring and analyzing errors related to conceptual, computational, and procedural mistakes. Mitigating these errors can help in the self-correction of generated answers, providing smarter future responses for the solution, valuable data for ongoing refinement, and insight into the AI’s ability to pinpoint areas where students commonly struggle.

The AI tutor’s impact and a shift in learning 

Through combining advanced reasoning, custom AI, and flexible data retrieval, the AI tutor yielded a remarkable 90% accuracy when addressing student queries across an average of 500 exam preparation questions. Essentially, this level of precision means students can confidently rely on the AI tutor for the majority of their questions.

But more than just accuracy, this technology is driving a fundamental shift in education. Students are now able to move beyond rote memorization and learn at their own pace through the AI tutor’s detailed, comprehensive responses. 

As an example, imagine a student struggling with a difficult mathematics problem late at night. Instead of getting stuck for hours, they can ask the AI tutor, giving them not only the correct answer, but a foundational understanding of the basic principles and the steps to get there.

This personalized, always-on exam companion continues to encourage independent thinking and understanding, thanks to a tech-first approach to education, and Infinity Learn’s collaborative work with Google Cloud Consulting.

“Since implementing the AI tutor, we’ve seen a remarkable shift in how students approach complex problems. They’re engaging with the underlying concepts on a deeper level. Google Cloud Consulting’s expertise was instrumental in helping us build a truly innovative AI tutor that goes beyond surface-level answers and cultivates real understanding.” – Amit Bansal, Chief Product Officer, Infinity Learn

Inspired by Infinity Learn’s transformative approach? 

The partnership between Google Cloud and Infinity Learn by Sri Chaitanya is just one of the latest examples of how we’re providing AI-powered solutions to solve complex problems and help organizations drive desired outcomes. 

Discover how Google Cloud Consulting can help you leverage Vertex AI to transform your own challenges or business needs. The Google Cloud Consulting portfolio provides unified services, bringing together offerings throughout multiple specializations like technical account management, professional services, customer success, and more. Check out Google Cloud Consulting’s full portfolio of offerings, or learn how to get started with Vertex AI.