AgroSentry

Inspiration

The FAO data shows an enormous amount of annual global food crops lost due to pests. 40% and a 220 billion USD in losses. This issue is much more important to solve than it seems. Therefore, we invent AgroSentry.

What it does

Basically, AgroSentry is a website that targets farmers as our customer. This website provides pest detection by image recognition of LLM, where it has been trained with the PlantVillage dataset of 320x320 resolution. Moreover, a Grad-CAM heatmap provides a visualization of regions that affect the model's decision-making most. With the data from journals and the ministry of agriculture, the OpenAI API is used to provide RAG to guide the farmers. Additionally, this website will be firmer, as there are more customers that use this website through anonymous community contribution and community forums. Each diagnosis contributes to anonymous metadata, and the forums enable farmers to interact with each other and share some knowledge and information.

How we built it

First of all, we brainstorm the ideas before the hackathon. At first, we wanted to choose the safety track, but then we changed our mind to agriculture. Secondly, we try to learn how to implement and develop the real product. We use several new coding languages to implement this website.

Challenges we ran into

As we have no experience of joining a hackathon before, we struggled to manage our time. Moreover, we are also newcomers that are interested in coding. Therefore, we learned a lot through our difficulties. We start from scratch, where we are not aware of how to implement FastAPI, ONNX and Tailwind. Through this hackathon, we could find our way to solve the challenges we ran into and be a better developers.

Accomplishments that we're proud of

We are proud of being able to code with new languages that we have not used before this competition. Moreover, we are proud to enrich our knowledge and push ourselves to the limits due to the time constraint.

What we learned

For the backend, we mainly learned about FastAPI because it is our main tool. Even though our application is still a little bit simple, we become more engaged about all of the aspects as we think and use it for our project. We have a past experience of coding, but it is only using a single programming language, but this FastAPI opens up our eyes to how components of web apps could interact with each other. Besides FastAPI, we learned so much about ONNX and RAG using the LangChain library. These aspects are a little bit time-consuming to learn about, but eventually, through the use of AI and YouTube, we could understand how to use those to make our MVP. For the frontend, we rarely used TypeScript for the frontend of our past project. Through this hackathon, we recall our skills and even enrich our skills regarding this UI/UX. We haven't known anything about TailWind before this hackathon. At first, we think that this is not different from CSS, but actually this is so much easier and simpler to implement. This is pretty useful for details of a website such as animation. By learning from doing, we could practically know how to implement it and would try to implement it in our future projects. After all, the main thing that we learned is from the pre-hackathon until the end. We learned how to brainstorm effectively with teammates, collaborate, network, and solve issues that people think are not important. More importantly, how to integrate the front-end and back-end. We assigned our jobs differently, which are front-end developer, back-end developer, and ML, database, idea developer. As an addition, we learned so much about time management. We usually make a project in the long term, but in this hackathon, we could learn how to develop a product, specifically an MVP, throughout 30 hours of coding.

What's next for AgroSentry

A live interactive disease map for farmers to see the distribution of diseases in a specific area and the distribution of a specific disease.

Built With

• fastapi
• next.js
• onnx
• rag
• supabase
• tailwind