This blog is written by Gokul Kumar from Madhi Foundation.
I have always looked forward to Tech4Dev’s meetups and community gatherings. I strongly align with their mission, which is the belief that communities solve problems and that technology simply acts as an enabler that brings these communities together. With that mindset, I was excited to join the AI Cohort 2.0 along with Ashwin, our Solutions Architect.
Even before the cohort began, our mentor had already been assigned, and we had started thinking about the direction we wanted to take. At Madhi Foundation, we have been building VallamAI, a mobile application designed to strengthen teacher coaching and support teachers and instructional coaches in improving classroom practices.
As part of the cohort, we were exploring three potential AI use cases:
- Speech recognition to analyse classroom conversations and teacher practices
- OCR for worksheets to digitise and interpret student responses
- A contextualised AI chatbot that could support teachers with relevant guidance
We entered the cohort with these ideas, but what followed over the next few days reshaped how we thought about them.
One of the early sessions focused on building guardrails for AI systems. As the session was happening, Ashwin and I began testing our chatbot in parallel. Almost immediately, we realized how many aspects we had overlooked, particularly around guardrails, reliability, and the ways an AI system could behave unexpectedly in real-world educational contexts. That moment became an important point where we decided that we would prioritise the chatbot for the next couple of months.
Over the next few hours, what truly stood out was the strength of the community.
“The moment we shared our broader problem statement, it stopped being just our problem.”
People from across organizations, along with our mentor and members from the Tech4Dev team like Rajagopal and Ashana, began brainstorming with us, asking questions, challenging assumptions, and suggesting approaches. What could have been an isolated technical exploration instead became a collective problem-solving process.
Equally valuable was learning from other organizations in the cohort, understanding the problems they were tackling, the AI approaches they were experimenting with, and the lessons they had already discovered. These exchanges created an environment where knowledge flowed freely rather than staying within individual teams.
But perhaps the biggest value add from the cohort was something simple yet rare in the social sector: the space to experiment. Many of us work in environments where immediate delivery often takes priority over exploration. The cohort created the time and environment for us to sit together, test ideas, break things, and iterate, something that is often difficult to carve out in our day-to-day work.
As we move forward over the next few months, our focus will be on strengthening the VallamAI chatbot. We are working toward:
- Improving contextualization, so the chatbot responds differently depending on the teacher’s location and context
- Building stronger guardrails to ensure responsible and predictable behaviour
- Increasing the accuracy and reliability of responses
- Understanding how an AI assistant can meaningfully support teachers’ decision-making processes
Ultimately, the goal is not to build AI for its own sake. The goal is to explore how AI, when designed thoughtfully, can support teachers in the complex work they do every day. And if there’s one thing this cohort reinforced for me, it is this:
AI may be the technology we are experimenting with, but community is still the real engine that drives meaningful innovation.
