This page of AgriScope offers various language options.
User inputs district, state and country here
App provides temperature, rainfall, humidity and soil pH according to the location provided by the user
User can either upload the picture of the crop or can describe the symptoms in words
User has to provide the month
User has to provide the budget for remedies
App provides the causes, remedies and further precautions of the that the user can take in the user's preferred language
App also provides the information regarding nearest KVKs that the user can visit and the documents that he or she needs to carry

Inspiration

India is an agrarian economy, yet millions of farmers lose significant portions of their yield to crop diseases due to a lack of timely expert advice. While advanced AI models can diagnose plants, existing solutions are often trapped behind language barriers, complex UI, or generic recommendations that ignore a farmer's tight budget. We wanted to build a practical, empathetic "AI Crop Doctor" that speaks the farmer's language, respects their financial constraints, and provides a direct, actionable bridge to physical agricultural centers like Krishi Vigyan Kendras (KVK).

# How We Built It

We developed our application using Google AI Studio and Gemini models to handle multimodal inputs (both images and text symptoms). Multimodal Analysis: The model processes real-time crop photos or text-based symptom descriptions to identify anomalies.

Dynamic Language & Context Adaptation: The system initiates a conversational flow that locks into the user's regional language and extracts geographical coordinates to estimate vital environmental factors.

Algorithmic Filtering: It maps the remedies against a user-defined budget threshold, optimizing for cost-effectiveness.

#What We Learned

Building this project taught us how to design conversational AI for users who require strict step-by-step guidance rather than overwhelming walls of text. We gained deep insights into prompt engineering, particularly how to enforce state-machine logic within an LLM system prompt. On the technical side, we explored how environmental metrics—such as regional temperature variations, rainfall patterns, and optimal soil chemistry—correlate with rapid pathogen spread.

#Challenges We Faced

The biggest hurdle was maintaining strict turn-by-turn conversational flow without the model rushing ahead, hallucinating user inputs, or triggering API timeouts. We initially ran into JSON parsing errors when trying to structure outputs dynamically across multiple steps. To fix this, we refined our prompt architecture to isolate instructions step-by-step, ensuring the model halts and waits for user input after every query. Another challenge was anchoring the AI's recommendations to real-world local administration parameters, which we resolved by strictly defining context guidelines for geographic mapping.

What's Next for AgriScope

Moving forward, AgriScope will transition from a standalone conversational prototype to a fully integrated ecosystem. Our immediate roadmap focuses on integrating live agricultural IoT sensors and weather APIs to replace estimations with real-time, precise environmental data tracking. We plan to build out localized database integrations to dynamically pull exact, verified contact details for KVK centers across all districts. Furthermore, we intend to expand the application's offline capabilities using lightweight, on-device models to ensure farmers can diagnose crop health even in low-connectivity rural zones, ultimately making AgriScope a ubiquitous, accessible tool for sustainable farming.

Built With

gemini-api
google-gemini-ai-studio
json
llm
react
typescript
vite

Submitted to

USAII® Global AI Hackathon 2026

Created by

As a key supporting member of the AgriScope team, I was responsible for enhancing the platform’s usability, visual identity, and overall user experience. My contributions include:

UI/UX Design & User Flow: Designed intuitive user interfaces and optimized screen layouts to ensure a simple, farmer-friendly experience across the application. Focused on accessibility and clear navigation to make interactions seamless.
Visual Branding & Prototyping: Developed the visual elements, color schemes, icons, and presentation materials that established AgriScope’s identity and improved user engagement.
Content & Media Production: Created and edited promotional videos, demos, graphics, and other multimedia assets used for project presentations, documentation, and outreach.
User Experience Optimization: Collaborated on refining conversational flows and interface interactions to improve clarity, readability, and ease of use for users with varying levels of digital literacy.
Presentation & Documentation: Prepared pitch decks, project demonstrations, and supporting documentation to effectively communicate AgriScope’s features and impact during hackathons and evaluations.
Testing & Feedback Integration: Assisted in usability testing, gathered feedback, and contributed to iterative improvements in the application's design and overall experience.
Team Collaboration: Worked closely with the primary developer to align technical capabilities with user-centric design principles, ensuring that AgriScope remained both functional and accessible for farmers.

Saikat Maity
As the primary developer of AgriScope, I spearheaded the technical architecture, prompt design, and core conversational flow. My specific contributions include:
System Architecture & State-Machine Design: Designed and engineered the core prompt engineering architecture within Google AI Studio. I built strict state-machine logic to ensure a stable, step-by-step conversational intake flow, effectively eliminating common LLM pitfalls such as API timeouts, hallucinations, and multi-turn JSON parsing errors.
Multimodal Integration & Adaptation: Implemented the integration of Gemini's multimodal capabilities, enabling the system to seamlessly switch between analyzing text-based crop symptoms and computer vision data from uploaded plant photographs.
Localization & Context Modeling: Developed the contextual logic that locks the conversation into the user's preferred regional language and maps user-submitted geographic locations to synthetic baseline environmental metrics (temperature, rainfall, humidity, and soil pH).
Algorithmic Constraint Filtering: Coded the logic boundaries that dynamically cross-reference potential plant remedies against the user's explicit budget constraints, ensuring the output prioritizes low-cost, accessible treatments.
Responsible AI & Resource Mapping: Formulated the negative constraints and safety guardrails to prevent unverified chemical generation, while structuring the final data output into an actionable, localized checklist for physical Krishi Vigyan Kendra (KVK) office navigation.

Agamani Mukherjee

Updates

Saikat Maity started this project — Jun 21, 2026 09:17 PM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.