AgriSmart
How can AI help farmers spot plant diseases and get reliable advice?
In partnership with
For the Ministry of Agriculture of Guinea
An academic project carried out with the startup INNOVUP, for the Ministry of Agriculture of Guinea. A smart-farming platform: an AI chatbot that advises farmers and diagnoses a disease from a photo, connected sensors in the field, a web app and a Flutter mobile app. I took part in designing the AI, the embedded part and the mobile app.
- My role
- Took part in designing the AI, the embedded part and the Flutter mobile app
- Duration
- 5 months (Jan – May 2026)
- Platforms
- AI, Web, IoT
- Stack
- LangGraph
- Python
- Flutter
- ESP32-CAM
- PyTorch
- RAG
- Spring Boot
- Angular
- Raspberry Pi
- MQTT / MQTTS
- FastAPI
- MongoDB
- FAISS
- ONNX
- TensorFlow
- LangChain
- Groq API
- MCP (Model Context Protocol)
- Machine learning
- Computer vision
- WebSocket
- Prompt engineering
Situation
38plant disease classes recognised by the ResNet9 model
Farmers often spot a disease too late, and rarely have an expert at hand. AgriSmart gives them an adviser available at any time, connected to their fields’ data.
Academic team project, carried out with the startup INNOVUP and built for the Ministry of Agriculture of Guinea, from January to May 2026. My role: taking part in designing the AI, the embedded part and the Flutter mobile app.
Product
- Photo diagnosis: send a leaf picture to the chatbot, it identifies the disease and suggests treatment and prevention.
- A chatbot that knows the farm: it answers from an agricultural knowledge base and the fields’ real data.
- Live sensors and anomaly alerts pushed in real time.
- Web (Angular) and mobile (Flutter), with a voice mode and a marketplace for products.
System
As the project is carried out with a partner, the detailed architecture stays confidential. Here is the stack.
- AI: Python, LangGraph, LangChain, LLM via Groq, RAG (FAISS), PyTorch, ONNX, TensorFlow.
- Backend: Spring Boot and Spring Cloud, FastAPI, Flask, MongoDB.
- Web and mobile: Angular, Flutter.
- Embedded: ESP32, Raspberry Pi, MQTT.
Model & data
- A multi-agent chatbot built with LangGraph, backed by an LLM and a search in an agricultural knowledge base (RAG).
- Computer vision: a ResNet9 model recognises 38 plant diseases from a photo.
- On-device AI: a first analysis runs directly on the sensors, in the field.