This 19-minute talk explains how I returned to an old agricultural product idea and built Krishi AI with a new generation of LLM-assisted tools. It is a practical account of what AI accelerated—and what still required product judgment, domain context, testing, and persistence.

What the talk answers

The talk connects three parts of my story: growing up in a farming family, building the earlier DigitalMandi app, and later working across software delivery, cloud, machine learning, and LLM systems. Krishi AI became the point where those threads met.

I describe how LLM tools supported research, ideation, architecture, coding, and debugging while I developed a real Android product. The central lesson is not that AI removes the difficult work. It changes the build loop: ideas can be tested faster, but the builder still has to understand the user, evaluate the result, fix failures, and keep the product alive after the first release.

Product themes covered

Moving from a historical farming-information app to a maintained AI-native product.
Combining AI chat, plant diagnosis, mandi prices, agricultural knowledge, and multilingual access.
Using LLMs across the product-development workflow rather than only as a user-facing chat feature.
The practical gap between a convincing prototype and a dependable public application.
Why farming context and continued user learning matter alongside technical capability.

Source and transcript status

The complete recording is available on YouTube and loads on this page only after you choose to play it. YouTube exposes automatic captions for the recording, but I have not published them here because a reviewed transcript is not yet available. The summary above is based on the talk’s public description and my published event notes.

SourceWatch on YouTube