Listen to this story
ChatGPT is redefining tech disruption. Today, developers are building different products on top of OpenAI’s GPT models. While there is an array of GPT-powered products and services, the one that caught our attention is KissanGPT. Developed by Pratik Desai, founder of Titodi, KissanGPT is an AI voice assistant for agriculture-related queries. Despite being a relatively new tool, it has already piqued the interest of the Indian government and other industry players.
KissanGPT leverages the power of GPT3.5 and Whisper models and has a knowledge layer on top of it. To keep the data up-to-date, Desai is constantly partnering with agricultural universities in the country. The platform comes with a voice interface, and currently supports nine Indic languages, including Gujarati, Marathi, Tamil, Telugu, Kannada, Malayalam, Punjabi, Bangla, and Hindi. Pretty soon, two more languages – Assamese and Odia – are expected to be added to the list.
Coming from an agricultural family, Desai decided to leverage his understanding of the sector. He has a PhD from the Wright State University and has been building AI/ML applications for agriculture for quite some time. Previously, he developed an automated labelling mechanism using generative AI. “Using Stable Diffusion within the first month of its release, we created nearly 20,000 stock images,” Desai said.
Besides, he has also created other tools to help farmers. Based in the US, Desai wanted to do something for Indian farmers back home. This is how KissanGPT was conceived. “We’ve been working in this space for some time, we have enough data in agriculture, and we are working with a farmer very closely,” he said.
One thing that Desai noticed is that even though there are many information channels in the agriculture domain, these channels are broken. For instance, farmers can seek information either from Kisan Call Centre, or through the internet. However, often farmers struggle searching for information online as accessing Google can be a challenge. Besides, the call centres only provide support for limited hours.
While there are different apps developed by different startups that provide farmers with information, Desai believes that the data provided is very fragmented, and at the end of the day, they want to sell their products. “We realised that while we are creating AI/ML applications, this is something we should be able to solve with the amount of data that we have,” he said.
With support from OpenAI and access to their models, Desai believes a tool like KissanGPT carries immense potential.
Unleashing the Power of LLMs
Desai wants to take his product, KissanGPT, to each farmer in every nook and corner of the world. However, he does agree that scaling is a challenge. Currently, he operates with a team of six and is funding the project from his own pocket. Having said that, the platform is growing organically and gets around 30,000 to 40,000 queries each month.
Desai also reveals an interesting development in Chhattisgarh, where a farmer, who came across KissanGPT, started educating his peers about it. “I was invited to one of their clubhouse, attended by around 450 farmers who had actually tried out our application,” he said.
Desai also plans to launch an application soon because he found that people are not very good at remembering URLs. “Once we have the app, we expect more and more farmers to use KissanGPT.”
Breaking Language Barriers
In India, farmers are spread in every corner of the country, and most of them have their own dialect. Even though Hindi is widely spoken in India, it might not necessarily be their mother tongue. For this, Desai has revealed that he is relying on the Indic language datasets and the IndicTrans translation model from Bhashini, an initiative by IIT Madras. “We have to make sure that the model we’re using is working for that language in terms of translation as well as transliteration and also text-to-speech in all parts,” Desai said.
The convergence of agriculture and AI is not novel. In fact, different parties have been leveraging AI to provide real-time information to farmers related to crops, soil, fertilisers and weather conditions.
However, “as far as I know, none of them comes with an end-to-end voice interface for Indian languages”, Desai continued, “that is a unique point, because we figured out that farmers actually like the voice interface which is much easier”. Secondly, Desai believes the dataset that KissanGPT has been trained on is one of his biggest assets. “Converting the data into something that works and providing information is something very unique to us.”
While Desai also has access to GPT-4, KissanGPT is powered by GPT3.5, because GPT-4 is expensive in comparison to the other LLMs from OpenAI. So, his goal is to make KissanGPT as inexpensive as possible. But, with GPT-4 with multimodal capabilities, it could potentially open up more use cases, especially with images.
However, he does acknowledge that Indian startups operating in this space could potentially build something similar. In fact, Desai has revealed that he is already in talks with one of these firms to integrate KissanGPT to their platform.