TEAM 14 – AGROSAARTHI
Empowering Indian farmers with portable and offline AI
Here’s the problem we couldn’t ignore:
AI has taken over almost every field, from finance to healthcare to education to transportation, bringing in unimaginable levels of efficiency and automation.
But when it comes to actual farming on the field, AI is still not widely used.
The main reason is that most farmlands, especially in India, are located in remote areas with poor or no internet connectivity. Most AI solutions today rely on fast internet and cloud-based computing to function effectively, making them out of reach for the majority of farmers.
This digital gap is exactly what we wanted to address! We asked ourselves, “why should our farmers be left out of the AI revolution just because of a lack of internet access?”
And that is exactly why we built AgroSaarthi! A smart, portable, and fully offline solution powered by TinyML that brings AI directly into the hands of our farmers. No apps, no Wi-Fi, just instant, on-field support exactly when it’s needed.
What got us started?
When we got into the Annam.AI Hackathon, we expected coding, brainstorming, and sleepless nights. But what we didn’t expect was how much we’d learn about real-world farming problems in just the first five days.
It all began with a series of talks by on-field research experts that gave us deep insights into the everyday challenges faced by Indian farmers, covering everything from sowing and harvesting to storage and distribution in great detail.
One idea stood out to us.
“What if farmers had a tiny, portable device that could help them understand and treat their soil and crops on the spot, instantly, without needing internet, apps, or guesswork?”
Most Indian farmlands do not have reliable internet access. Farmers need offline, easy-to-use tools to solve problems directly in the field.
And that became our mission!
The making of AgroSaarthi:
We kicked things off by learning the basics of farming and understanding what makes soil healthy. We studied key parameters like pH, NPK values, moisture levels, and temperature. We understood how different soil parameters affect each type of crop growth. We also looked into common pest issues and nutrient deficiencies across different crops.
Based on this, we built two core machine learning models:
- Soil Health Model
- Pest Detection Model
Our soil health model was trained using the ideal soil parameters for different crop types. When a farmer selects a crop, the model compares real-time soil readings against these standards and provides tailored recommendations. It’s like having a crop expert right there in the field, giving advice on the spot!
We trained and converted both models into TinyML using the ONNX format, and deployed them on a Raspberry Pi 5.
For the hardware, we connected a RS485 5-in-1 soil sensor that measures pH, temperature, and NPK levels, along with a camera module for pest detection. All the recommendations were displayed on an onboard OLED screen.
The entire setup was designed to work completely offline. A self-contained, field-ready tool we named AgroSaarthi.
Throughout the process, we regularly had meetings and discussions with our college mentor, Dr. Sundharakumar, whose guidance helped us refine both the technical and practical aspects of our solution. We also received expert mentorship from Mr. Soma Dhavala, who was introduced to us through Annam.AI, especially for navigating TinyML. In addition, Dr. Suman and his intern team were supportive during the hardware phase, assisting us in solving key integration issues.
The Not-So-Fun Part (Challenges)
Like any hardware-software project, we faced a fair number of challenges.
One of the biggest was the lack of clean, reliable soil health data from Indian sources. This made it difficult to train our models for a real world scenario.
Another major learning curve was TinyML itself. It was a relatively new and unexplored field for us, and working with it was a completely different experience compared to traditional ML development.
Compressing machine learning models to run on a RPI5 without compromising on accuracy was tricky. Finding the right balance between model size and performance took a lot of trial and error.
We also received the hardware components fairly late into the hackathon timeline. This led to a rush during the final stages, with unexpected compatibility issues between sensors and the Raspberry Pi. Testing everything under time pressure brought its own set of frustrations.
But all that last-minute stress was completely worth it when we finally saw our prototype come to life and work efficiently in real time. Seeing AgroSaarthi run exactly the way we imagined made it all incredibly satisfying!
Support from the Annam.AI Team
We truly appreciate the Annam.AI team for providing us with a seamless hackathon experience. They were always available to address our queries and offered guidance whenever we needed it. When we mentioned that TinyML was new to us, they connected us with an expert in the field, which was a huge boost to our learning curve.
Their support didn’t stop there. They also helped us with reimbursement for the purchase of hardware components, and a team from IIT Ropar couriered us a Raspberry Pi 5 for our prototype development. Their encouragement and timely assistance played a crucial role in shaping AgroSaarthi.
What’s Next?
AgroSaarthi is just the beginning. We are already exploring features for the next version, such as:
- Audio-based pest detection
- Solar-powered battery support
- Multilingual voice feedback
- A companion mobile app for offline syncing and record-keeping
Our goal is to keep improving this device and make it even more accessible, efficient, and practical for farmers across India.
“Technology should feel natural and useful, not complicated. And that’s exactly what we want AgroSaarthi to stand for.”
The Team:
- Varsha Pillai M
- Vikranth V
- S Nittin Balajee
- Lithikha B
We’re third-year students from Shiv Nadar University, Chennai, who love building tech that matters. AgroSaarthi was our first real hands-on project, and it took us on a wild ride through AI, hardware, and a whole lot of farming knowledge.
Thank you for reading!