Gemination – Simple Guide

This project is called Gemination.
It is an agriculture helper app that uses Google Gemini 3.

With this app, a farmer can:

• Take a photo or video of a crop
• (Optionally) record audio to describe the problem
• Send it to the app
• Get a diagnosis: crop, possible disease, treatment, and a confidence score
• Ask follow‑up questions in a chat

If the AI is not very confident, it can search the web to double‑check the answer.
If you send your location (lat/lng), the app also uses where you are to adjust the diagnosis.

There is:

• a backend API in the api folder (FastAPI + Gemini 3)
• a frontend web app in the web folder (Next.js/React)

---

How to Run the Backend (API)

1. Open a terminal in the api folder

   • Example:
     Go to the project folder, then:
     • cd api

2. Create and activate a Python virtual environment (only needed once)

   • On Windows (PowerShell):

     python -m venv venv
     .\venv\Scripts\activate

3. Install Python packages

   pip install -r requirements.txt

4. Set your Gemini API key

   • Create a file called .env inside the api folder.
   • Put your key inside like this:

     GEMINI_API_KEY=your_key_here
     GOOGLE_CLOUD_CREDENTIALS_JSON=your json_content_here
     

5. Start the backend server

   uvicorn main:app --reload

   • The API will usually run at http://localhost:8000.
   • You can see the docs at http://localhost:8000/docs.

What the Backend Does

• User accounts: create user, login, get current user (authentication).
• Diagnosis:
  • Receive image, audio, and video.
  • Save files in static_image, static_audio, and static_video.
  • Call Gemini 3 with the media and optional location.
  • Get back: crop, disease, treatment, and confidence.
  • If confidence is low, call Gemini again with web search to verify.
  • Store the diagnosis in the database.
• Chat:
  • Let users ask questions about a specific diagnosis.
  • Use Gemini 3 to answer.
  • Use the diagnosis info, chat history, and location context.
  • Follow confidence rules (say “unsure” when confidence is low).
• Translation:
  • Translate between the user’s language and English.
  • This is used both for the diagnosis and for chat replies.

---

How to Run the Frontend (Web App)

1. Open a second terminal in the web folder

   • From the main project folder:

     cd web

2. Install Node packages (only needed once)

   npm install

3. Set up .env.local environment

   • Create a file called .env.local file inside
   • Place the backend local host like this:

     NEXT_PUBLIC_API_URL=http://localhost:8000

4. Start the dev server

   npm run dev

5. Open the app in the browser

   • Go to http://localhost:3000

What the Frontend Does

• Auth pages
  • Register: create a new user (this calls the backend user API).
  • Login: log in the user and store the token.
• Main agriculture flow
  • Page to upload images / video of crops.
  • Attach audio (or use audio that was transcribed).
  • Send location (lat/lng) from the browser if allowed.
  • Show the diagnosis result from the backend: crop, disease, treatment, confidence.
• Chat page
  • Let the user ask questions about a chosen diagnosis.
  • Show AI answers and confidence flag (high/low).

How Everything Fits Together

• The web app talks to the backend API using HTTP (usually localhost:8000).
• The backend API talks to Gemini 3 using your GEMINI_API_KEY.
• The database stores:
  • users
  • diagnoses
  • chat history
• Location + media + text all work together to give better crop diagnostics.

If you follow the steps above (backend first, then frontend), you should be able to:

1. Register a user
2. Log in
3. Upload crop media and location
4. See a diagnosis and confidence
5. Chat with the AI about that diagnosis