Meta Model API

CASE STUDY

Giving students an AI advisor for faster, better research on higher education

At a glance

Industry

Technology

Use case

Enabling chat-based college and course research for students

Goal

Transform decision-making about higher education into a faster, more tailored experience

Llama versions

Llama 3.1 8B

Deployment

On-premises

Download full case study

THEIR STORY

Helping students find their path forward

Info Edge India Limited’s online platform, Shiksha.com, empowers students to make informed education and career decisions. The platform provides them with credible, updated and personalized information about colleges, courses, exams and career paths in India and abroad.

THEIR GOAL

Speed up and personalize the college research process

Students in India and abroad struggled to find reliable college, course and exam information, navigating scattered sources and tools that didn’t understand the context of their questions. Shiksha.com, part of Info Edge, was solving the problem for millions of students with a single site for credible information. But they wanted to use generative AI to make the college research process faster and more personalized. Their goal was to build a conversational chat solution that could deliver precise, real-time answers while keeping performance high and costs manageable as they scaled.

THEIR SOLUTION

A scalable and cost-efficient Llama-enabled chatbot for students

The team built ShikshaGPT, an AI chatbot for college research, using lightweight Llama 3.1 8B. After testing multiple models, they selected Llama 3.1 8B for its strong balance of performance and cost, avoiding the high compute demands and rising inference fees of larger closed models. The model’s compact size allowed Shiksha to deploy efficiently on premises, ensuring data control, security and scalability. Llama 3.1 8B delivered fast inference times, handled multiple users at once without issue and demonstrated proficiency with academic language. Plus, its open-source nature gave the team flexibility to deploy and iterate quickly.

THEIR APPROACH

Delivering accurate, personalized responses with RAG

With the Llama 3.1 8B model and a retrieval-augmented generation (RAG) workflow, ShikshaGPT answers complex education questions using curated information from academic databases, syllabi and more. Before reaching the model, each query goes through three pre-processing steps:

Hinglish correction to standardize language in both English and Hindi
Entity recognition to extract relevant academic terms
Intent classification to determine the query’s purpose

THEIR SUCCESS

Supporting faster and better college research at scale

ShikshaGPT has empowered Shiksha.com to deliver a dynamic AI advisor that students trust to help them improve their decision-making about colleges and courses. The chatbot also drives a measurable business impact, encouraging students to take actions that confirm their intent to apply — a metric tied to revenue growth. With Llama 3.1 8B, the team is able to scale the solution to serve more students at a fraction of traditional costs.

85% net positive user feedback on the quality of ShikshaGPT answers
10% of the mobile app’s responses earned “intent to apply” actions from users

Please confirm your identity