The problem कृषिबंधू (Krishi-Bandhu) solves

Current Scenarion

• Agriculture is the backbone of Indian Economy, employing more than half of the country’s population.
• Alone 2/3rds of world’s food production is supplied by smallholder farmers having land less than 5 acres and they constitute 86% of all the farmers
• Even though scientific and technological innovations and political reforms over time have fueled this primary occupation, smallholder farmers face significant challenges in adopting sustainable practices.
• Limited access to real-time data and its timely analysis hinders their efficient farming cycle.
• Moreover, no such integrated application exists that can help them with efficient resource management techniques and training, give accurate information on daily weather, give assistance over crop disease, and more importantly help them with the required knowledge to increase their cultivation generating quality yield.

Solution

Through ‘कृषि बंधू,' we aim to empower smallholder farmers with advanced technological tools. 'कृषि बंधू' addresses critical use cases -

• crop yield prediction through machine learning
• early crop health diagnostics with AI
• decision support systems for climate adaptation
• multilingual chatbot with speech-to-text that educates farmers on agriculture, crop queries, government schemes, and comparative market crop rates for informed decisions
• send weather updates and alerts via SMS and calls (offline), ensuring farmers can prepare for adverse conditions and minimize crop damage

This comprehensive approach helps farmers make data-driven decisions, manage resources efficiently, and adapt to climate change, ensuring a sustainable agricultural future.

Challenges we ran into

Here’s a breakdown of specific challenges and solutions we encountered while building the project:

Offline Execution of ML Model for Plant Disease Detection

Challenge: We implemented an ML model for plant disease detection, which ideally needed to run offline on the client-side. However, running a model locally can be challenging due to memory constraints and the need for fast inference, especially on devices with limited computational power.

Solution:

1. Model Optimization: We utilized TensorFlow.js, a JavaScript library for running ML models directly in the browser. However, the model was initially too large for efficient inference on some devices, causing lags.
2. Progressive Loading: To further optimize, we implemented a progressive loading approach that loads only essential parts of the model at once, allowing it to start running without having to load the entire model initially.

This process required testing various configurations to balance model size, accuracy, and speed. Ultimately, the optimizations allowed for a smooth user experience even offline.

Integrating Gemini API for Chatbot

Challenge: Integrating the Gemini API to power our chatbot functionality presented challenges with response speed and API rate limits.

Solution: Thanks to AI studio for generating Gemini API keys for free powering our chatbot