# Project Atharv | Sherwin Vishesh Jathanna

> Launched Atharv, an AI-powered agricultural assistant. Gained 5,000 users in 48 hours. Presented it to the Ministry of Agriculture for the state of Karnataka a…

Source: https://www.sherwinvishesh.com/project/project-atharv  
Copyright 2026 Sherwin Vishesh Jathanna. Text may be quoted with attribution. The design and source code are not licensed for reuse.

Launched Atharv, an AI-powered agricultural assistant. Gained 5,000 users in 48 hours. Presented it to the Ministry of Agriculture for the state of Karnataka and was featured in 20+ news articles. Atharv improved crop disease diagnosis and weather prediction accuracy by 73%, providing personalized farming insights in local languages.

**Project Atharv** is the most ambitious thing I have built: an *AI-powered agricultural assistant* meant to put real, data-driven farming advice into the hands of the people who have always sat on the wrong side of the technology gap. From day one the goal was *accessibility over cleverness*. A farmer should be able to get a straight answer about their crop, in *their own language*, without needing English or a fast connection. The response was immediate. Within **48 hours of launch it had over 5,000 active users**, and it went on to be presented to the **Ministry of Agriculture for the state of Karnataka** and featured in more than **20 news publications**.

### The pivots that shaped it

Atharv did not arrive fully formed. It *started as a custom CNN*, a convolutional neural network I trained to recognize crop diseases from photographs. It worked in a controlled setting, but it kept hitting the same walls: it demanded enormous, cleanly labeled datasets and stayed brittle against the messy reality of real fields, different crops, uneven light, and cheap phone cameras. So I made the hard call to *drop the custom model and move to the Gemini API*, which generalized far better across crops and languages and freed the project to focus on being genuinely useful instead of endlessly retraining.

There was a second pivot too. I originally wanted Atharv to live inside **WhatsApp**, since that is the one app nearly every farmer already has and knows how to use, which made it feel like the most natural front door. In the end, practical constraints steered it toward a **web app** instead, and that is where it settled, still deliberately lightweight and *optimized for low-bandwidth areas* so it would work where connectivity is thin.

### What Atharv actually does

Atharv works through three simple tools. The **image-based chatbot** lets a farmer snap a *photo of a struggling plant* and get back a real diagnosis, not just of crop diseases but of *soil health issues* too, along with corrective steps and preventive measures. Because it runs on **Google's Gemini**, it is not boxed into a fixed list of a few dozen problems, it can reason across a *wide range* of crops and conditions and explain what to do about each, from treatments to fertilizer and pesticide guidance. The **text-based chatbot** answers almost any farming question in plain language, whether it is about crop rotation, fertilization techniques, or pest management. And **weather insights** turn a farmer's *pincode* into local, real-time forecasts. A farmer simply enters their zip code and gets the weather for exactly that location, drawn from **NASA's weather API** so the readings are far more accurate than a generic regional forecast, and Atharv translates that rainfall and temperature into concrete advice on when to plant, irrigate, and harvest.

The thing that lets all of this actually reach people is **language**. Atharv speaks in the farmer's *own* tongue, and thanks to **Google Translate** it is not limited to a handful of languages, it supports a *broad range of Indian and regional languages*, so the advice lands in whatever language a farmer is most comfortable with. Everything runs on a lightweight **Python and Flask** backend, with **Pillow** handling the uploaded crop images, deliberately built to stay usable on thin connections.

### Who really uses it, and how

Here is the honest, and to me most interesting, part of the story. Atharv eventually got **integrated into eSAP, the Karnataka state government's agriculture app**, which was a genuine milestone. But adoption did not unfold the way I first pictured it. Atharv never became the app sitting in every farmer's pocket. Instead, its *heaviest users turned out to be the people one link up the chain*: the **dealers who sell pesticides, fertilizers, and crop medicines**. Farmers have always walked into those shops to ask what is wrong with their crop and what to buy, and now those sellers pull up Atharv to answer them. So the tool reaches farmers *indirectly, through the agricultural supply chain*, and within that community of input retailers it is used heavily and often daily. That was not the path I predicted, but it is a real one, and it taught me that adoption tends to follow the habits people already have rather than the ones we design for them.

### More than an app: the people I met

Building Atharv meant getting off the screen and onto the ground. I *reached out to countless farmers*, sat with them in their fields, and worked to understand their problems before writing a single line of the solution. That effort opened doors I never expected: presenting to the **District Magistrate of Bengaluru**, the **Commissioner for Agriculture**, and the **Minister of Agriculture for Karnataka**, speaking at press conferences, and watching the project get covered across Kannada, Urdu, and regional media. Improving crop disease diagnosis and weather prediction accuracy by **73%** was the technical headline, but the part that stayed with me was the *people* this one project let me meet, the farmers, officials, journalists, and experts along the way. Atharv is, for me, proof that AI only means something when it actually reaches the human beings it was built for.

## At a glance

- **Timeline:** May 2024 - Aug 2024
- **Type:** Community Project
- **Recognition:** Featured in 20+ news articles, presented to Ministry of Agriculture

## Technologies

- Python
- Flask
- Google Gemini 1.5 Flash
- Google Translate
- NASA Weather API
- Pillow (PIL)
- Natural Language Processing (Multilingual Chatbot)
- Computer Vision
- Deep Learning (CNN)

## Categories

- Web App
- AI/ML
- Accessibility
- Farming

## Photos

- Presenting Project Atharv to the District Magistrate of Bengaluru
- With Dr. Thuppil Venkatesh, India's leading lead poisoning expert
- Handing the Atharv overview to Y.S. Patil, IAS, Commissioner for Agriculture
- At the Department of Agriculture, Government of Karnataka
- Presenting Project Atharv to the Minister of Agriculture, Karnataka
- Project Atharv featured in the Urdu daily Roznama Rashtriya Sahara
- Speaking to the press about Project Atharv
- At the Project Atharv press conference in Bengaluru
- Project Atharv covered in the regional Kannada press
- Project Atharv featured across the Karnataka news media
- Talking with a farmer about crop disease in her field
- Project Atharv featured with a banner headline in a Kannada daily
- Project Atharv covered in the Kannada press
- Understanding a farmer's needs on the ground

## Links

- [Linkedin Post](https://www.linkedin.com/feed/update/urn:li:activity:7234757502884532224/)