Zeal Education

Programs using AI

Over the last couple of years, we have realized that Artificial Intelligence can help solve many of the problems that we face in implementing programs in education: how to create multiple resources for the same theme, how to ensure that quality education reaches everyone, how to ensure that teachers can concentrate on teaching, etc. Our work is based on novel paradigms.

 

The Teacher in Charge

Take a simple problem like creating a presentation for a class. Let us assume that the teacher has an assistant who is an expert at the job. Typically, the teacher and the assistant will have a series of discussions, starting with the basic plans for the presentation, deciding on the contents of each slide, deciding on what media are appropriate and finally deciding on the layout, colour scheme, transitions, etc. In contrast, AI based apps, given a prompt, directly provide the final presentation!

The Teacher in Charge is based on the idea of designing apps with the same workflow. Given a prompt for a story, for example, first the app produces the basic stories, then the prompts for the images to illustrate the story, then produces the images themselves, then produces the voice over for the text and only after each stage is approved by the teacher, puts everything together to produce the final version. This way of writing apps is based on a simple idea:

The teacher knows her students and exactly what she wants, AI knows how to do it!

  • Kathakar: Create stories with illustrations and voice over
  • Prashna Mitra: Create worksheets with complete control over questions
  • Prastut Karta: Create presentations step by step

GAIA: Generate, Assess, Improve, Approve

One of the biggest issues with AI is the possibility of errors in the generated product: the teacher may specify that a question be appropriate for grade 6, the AI generated question may be appropriate for grade 7. The GAIA paradigm similifies the teacher's job by using two LLMs:
  1. The teacher gives a prompt, including e.g. topic, class, board, difficulty level, Bloom level and type of question.
  2. The prompt is given to LLM1 and it Generates a question.
  3. This question is sent to LLM2, which produces an Assessment report about what class, difficulty level, etc. the question is appropriate for. LLM2 is completely unaware of the teacher's requirements.
  4. The app compares the teacher's requirements with the report. If there are differences, the report is sent back to LLM1 for an Improvement in the question, a fresh report is generated by LLM2 and the process continues till a proper question has been generated.
  5. The app finally shows the question to the teacher for Approval. The Teacher gets not only a question that she can use in her worksheet, but also a certificate from LLM2 about the appropriateness of the question!

All Inclusive Artificial Intelligence

At the VivaTech 2026 summit in Paris, Prime Minister Narendra Modi emphasised that AI stands not only for Artificial Intelligence but also All Inclusive: to be truly useful, the results of AI must reach everyone! Our CSR work has exposed us to many of the problems that students in rural areas face and we have developed apps to ensure that AI reaches out to everyone! All the apps interact with the student exclusively in her language (Gujarati to begin with).

  • Abhyas Mitra: An app to help a rural student prepare for her examinations
  • Disha Mitra: An app to provide career guidance to students
  • Naukri Mitra: An app to provide interview preparation assistance for students.
  • Angrezi Seekho: An app to help rural students learn spoken English

Special Programs for the Visually Handicapped

One of the greatest issues that the visually handicapped face is access to material. Books in Braille are very expensive, heavy and often not available. While a normal reader has access to millions of pages of text, most of these remain inaccessible to the blind user. One solution that has been tried is to use refreshible Braille displays that take a text letter by letter, convert it into Braille and display it using retractible pins. These devices are extremely expensive and beyond the reach of the majority of India's blind population.

We have come up with two solutions, one using AI:

Jnana Vani: Information through sound. Text is converted to speech but AI is used to identify sentence and paragraph boundaries. The user uses a simple 5 button keyboard to navigate by paragraphs and sentences rather than by time!

Jnana Deepa: A mechanical Braille device which uses a series of embossed wheels that rotate to display characters in Braille. To display 20 characters, the device uses 20 motors rather than the 120 miniature relays needed to control the pins in refreshible displays!