Asaan Kisaan

Why Slide
Understanding the Challenge
Solution Overview Slide
Agentic Expert
Technical Implementation

🌾 About the Project — Asaan Kisaan

💡 Inspiration

Farming in Pakistan often suffers from poor internet connectivity and limited access to expert guidance.
Farmers struggle to diagnose plant diseases, interpret weather patterns, and track market prices.

We created Asaan Kisaan — an AI-powered farming assistant that works both offline and online, ensuring every farmer has access to instant, reliable insights.

⚙️ How We Built It

Our architecture blends on-device intelligence with cloud-based agentic reasoning.

Offline Mode:

Lightweight MobileNetV3 model for crop disease detection.
FAQ RAG for answering common farming questions locally.

Online Mode:

Multi-agent system with dedicated Crop, Weather, and Market experts.
Orchestrated using the Gemini API for intelligent routing between agents.
Backend powered by Firestore and GCP Cloud Functions for real-time data sync.

🧠 What We Learned

Building hybrid AI systems that adapt to connectivity conditions.
Implementing multi-agent reasoning pipelines.
Balancing accuracy vs. latency on resource-constrained devices.
Understanding real-world AI for social impact in rural ecosystems.

🚧 Challenges We Faced

Compressing the CV model for smooth mobile inference.
Designing dynamic agent triggers for different user intents.
Ensuring data privacy during disease detection.
Creating a minimalist UI that remains intuitive for non-tech users.

🌍 Impact

Asaan Kisaan provides every farmer with a personal AI assistant that diagnoses crop diseases, forecasts weather, and suggests market prices —
all while being accessible offline and online.

“Empowering farmers with AI, one harvest at a time.”

🧰 Tech Stack

Frontend: Kotlin
Backend: Firestore, GCP Cloud Functions
AI Models: MobileNetV3 (on-device), Gemini API (LLM)
Features: Offline detection, Multi-agent system, FAQ RAG