Situation. The company's AI tool was producing exam questions that sounded right but were often factually wrong. The output was confident and unreliable, so the team could not hand it to students without heavy manual review.

Action. I rebuilt the prompt architecture from scratch. Added tighter topical constraints, a layered self-check step, and output formatting rules that matched real exam patterns. Tested each version against known questions, iterated until the error rate dropped. Also designed the student workbooks and wrote quality protocols for the team. Reported to Rohini Saini.

Result. Over 2,000 high-accuracy board exam questions generated, along with a workflow the team could run without me. Quality protocols and workbook templates are still in use.

Learning. Reliability is a design problem. You have to build verification into the system, not bolt it on after. The hardest part was not generating good content. It was building a system that caught its own mistakes gracefully.

Situation. I was invited to join UNESCO's Education for Sustainable Development program, a global curriculum focused on systems thinking, long-term consequences, and responsible decision-making.

Action. Worked through every module across sustainable systems, community-centred development, and long-view thinking. Applied the frameworks to real questions I was already asking about entrepreneurship and community.

Result. Completed the program. Walked away with a new frame for thinking about impact, not just as a checkbox, but as a design constraint that sits at the centre of the work.

Learning. Sustainability and scale are not opposites. You plan for both from the start, or you plan for neither.

Situation. In Class 11, my classmates and I kept running into the same issue. We knew the content, but did not know how top CBSE scorers actually structured their answers. The format, the language, the presentation. No existing tool addressed this specifically.

Action. I built a Custom GPT trained on real CBSE topper answer scripts. Curated the training material, designed the prompts to extract structural patterns, and tested it against actual exam questions until the output genuinely matched topper-level framing.

Result. My friends started using it for exam prep and told me it changed how they wrote answers.

Learning. Specificity beats generality. A tool built for one exact problem outperforms a general chatbot every time.

Situation. During the COVID lockdown, school stopped and my friends and I were stuck at home. People wanted to learn things like guitar and drums, but no one knew where to start inside a sea of YouTube videos.

Action. I built a website that curated the best YouTube tutorials for specific skills in one organised place. Chose links based on how well they explained things for beginners, structured them by topic, and shared it with my circle.

Result. Friends started using it and, more importantly, started contributing their own finds. It grew beyond what I had planned.

Learning. Useful things spread on their own. If something genuinely solves a problem people already have, you do not need to market it.

Situation. Chapter Wise's AI pipeline was generating board exam questions that hallucinated. The prompts had no verification layer, and the team could not ship the output to students.

Action. Redesigned the prompt system from scratch. Added topic constraints, a self-check step, and exam-style formatting rules. Tested each iteration against known questions until the failure rate dropped to acceptable levels.

Result. Generated over 2,000 accurate board exam questions. Designed the accompanying student workbooks. Wrote the team's quality protocols.

Learning. AI tools only work when you design for failure as carefully as you design for success.

Situation. St. Xavier's World School asked me to run an AI workshop for students, most of whom had never used AI tools before. The risk was making it feel either intimidating or irrelevant. A lecture instead of something usable.

Action. Designed the workshop from scratch. Interactive demos, activities that needed no prior knowledge, and real examples tied to problems students actually cared about. Ran the session and adjusted based on how the room was responding.

Result. The school issued a formal appreciation letter. More importantly, students left able to use AI tools practically, not just as a novelty.

Learning. The gap between being taught and actually understanding is almost always a design problem. If something is not landing, the explanation needs to change, not the audience.

Situation. The company's AI tool was producing exam questions that sounded right but were often factually wrong. The output was confident and unreliable, so the team could not hand it to students without heavy manual review.

Action. I rebuilt the prompt architecture from scratch. Added tighter topical constraints, a layered self-check step, and output formatting rules that matched real exam patterns. Tested each version against known questions, iterated until the error rate dropped. Also designed the student workbooks and wrote quality protocols for the team. Reported to Rohini Saini.

Result. Over 2,000 high-accuracy board exam questions generated, along with a workflow the team could run without me. Quality protocols and workbook templates are still in use.

Learning. Reliability is a design problem. You have to build verification into the system, not bolt it on after. The hardest part was not generating good content. It was building a system that caught its own mistakes gracefully.

Situation. I was invited to join UNESCO's Education for Sustainable Development program, a global curriculum focused on systems thinking, long-term consequences, and responsible decision-making.

Action. Worked through every module across sustainable systems, community-centred development, and long-view thinking. Applied the frameworks to real questions I was already asking about entrepreneurship and community.

Result. Completed the program. Walked away with a new frame for thinking about impact, not just as a checkbox, but as a design constraint that sits at the centre of the work.

Learning. Sustainability and scale are not opposites. You plan for both from the start, or you plan for neither.

Situation. In Class 11, my classmates and I kept running into the same issue. We knew the content, but did not know how top CBSE scorers actually structured their answers. The format, the language, the presentation. No existing tool addressed this specifically.

Action. I built a Custom GPT trained on real CBSE topper answer scripts. Curated the training material, designed the prompts to extract structural patterns, and tested it against actual exam questions until the output genuinely matched topper-level framing.

Result. My friends started using it for exam prep and told me it changed how they wrote answers.

Learning. Specificity beats generality. A tool built for one exact problem outperforms a general chatbot every time.

Situation. During the COVID lockdown, school stopped and my friends and I were stuck at home. People wanted to learn things like guitar and drums, but no one knew where to start inside a sea of YouTube videos.

Action. I built a website that curated the best YouTube tutorials for specific skills in one organised place. Chose links based on how well they explained things for beginners, structured them by topic, and shared it with my circle.

Result. Friends started using it and, more importantly, started contributing their own finds. It grew beyond what I had planned.

Learning. Useful things spread on their own. If something genuinely solves a problem people already have, you do not need to market it.

Situation. Chapter Wise's AI pipeline was generating board exam questions that hallucinated. The prompts had no verification layer, and the team could not ship the output to students.

Action. Redesigned the prompt system from scratch. Added topic constraints, a self-check step, and exam-style formatting rules. Tested each iteration against known questions until the failure rate dropped to acceptable levels.

Result. Generated over 2,000 accurate board exam questions. Designed the accompanying student workbooks. Wrote the team's quality protocols.

Learning. AI tools only work when you design for failure as carefully as you design for success.

Situation. St. Xavier's World School asked me to run an AI workshop for students, most of whom had never used AI tools before. The risk was making it feel either intimidating or irrelevant. A lecture instead of something usable.

Action. Designed the workshop from scratch. Interactive demos, activities that needed no prior knowledge, and real examples tied to problems students actually cared about. Ran the session and adjusted based on how the room was responding.

Result. The school issued a formal appreciation letter. More importantly, students left able to use AI tools practically, not just as a novelty.

Learning. The gap between being taught and actually understanding is almost always a design problem. If something is not landing, the explanation needs to change, not the audience.