IIM Bangalore asks students to build chatbots to answer questions in midterm exam

IIM Bangalore students faced an unusual Generative AI midterm where they built their own chatbots to answer 15 questions. They wrote prompts, supplied course material as memory and tested multiple bots before submitting one for grading. The AI-assisted exam platform was developed by their professor and hosted on Vercel.

In Short

• IIM Bangalore students built their own chatbots instead of taking a written midterm exam
• Students supplied course material as memory and optimised prompts to answer 15 questions
• The professor built the exam platform using AI-assisted coding and hosted it on Vercel

What if your next college exam did not ask you to write answers, but to build an AI that could write them for you?

That is exactly what happened in a Generative AI course for Year 2 PGPBA students at IIM Bangalore.

Instead of sitting for a conventional pen-and-paper midterm, students were asked to create their own chatbots and use them to answer 15 exam questions.

EXAM WAS ABOUT MORE THAN AI

The assessment, designed by IIM Bangalore Decision Sciences faculty member Prof Subho Majumdar, required each student to configure a chatbot using a platform developed by the professor.

Students had to write their own prompt and provide course material and other relevant content as the chatbot's memory, as per a LinkedIn post by IIMB.

Their performance therefore depended not just on knowing the subject, but also on how effectively they could optimise the prompt and memory so their chatbot produced better answers.

The final responses were graded by the teaching assistant, much like a conventional exam.

PROFESSOR BUILT THE PLATFORM WITH AI

Majumdar wanted the assessment to better match the practical nature of a Generative AI course, rather than relying entirely on a format that could restrict how students engage with the material.

The platform itself was created using AI-assisted coding and hosted on Vercel. Each student received a separate login and could experiment with multiple chatbots before choosing one for submission.

In a course about Generative AI, the exam ended up becoming a practical AI exercise itself.

Students were not simply being asked whether they understood AI. They had to make AI work for them.

- Ends