TeaCareAI

TeaCareAI is a full-stack intelligence platform tailored for tea farmers and agronomists, specifically designed to address tea leaf diseases in Sri Lanka and beyond.

By taking an ordinary smartphone picture of a tea leaf, the application leverages deeply trained vision models and Retrieval-Augmented Generation (RAG) to precisely evaluate tea leaf health, provide real-time treatment guidance, and track the recovery progress of affected fields over time.

Key Features

1. Leaf Status Verification & Disease Detection

• Leaf Verifier (leaf_verifier.py): Pre-validates uploaded images to ensure they represent a valid tea leaf, minimizing false positives on stray photos.
• Disease & Severity Identification (tea_disease_identifier.py): Employs an Ultralytics YOLO (PyTorch) model (tea_disease_identifier_weight.pt) to detect the specific disease (e.g., Blister Blight, Pestalotiopsis) and calculate the percentage of leaf infection/severity.

2. Context-Aware Treatment Recommendations (RAG)

• RAG Engine (treatment_recommendations.py): Uses a local Vector Database (ChromaDB + SentenceTransformers BAAI/bge-small-en-v1.5) mapped to a localized treatment repository.
• Natural Language Explanations: Uses Ollama running llama3.1:8b right on the edge to turn agriculture knowledge into actionable advice for the farmer—working entirely offline if needed.

3. Analytics & Recovery Tracking

• Post-treatment Tracking (recovery_tracker.py): Associates scans using a unique chat/scan barcode. Uses a TensorFlow/Keras Neural Network (recovery_model.h5) to analyze sequential data and deduce whether a plant's health is "Improving," "Deteriorating," "Stable," or "New."
• Field Management: Geo-tags and tracks crop elevation, tea variety, and plant age for aggregate analytics across the property.

---

Architecture & Project Structure

The project relies on a modular Flask backend interfaced with independent machine learning services, bound together by a robust PostgreSQL data management tier.

TeaCareAI/
├── app.py                     # Flask application routes
├── main.py                    # Entry point: runs pre-flight constraints, then launches Flask
├── controller.py              # Business logic coordinating services & DB operations
├── auth.py                    # Session management and generic auth mechanisms
├── bootstrap.py               # Creates target DB schema & primes the LLM
├── check_constraints.py       # Ensures Python version, Ollama, & DB are active before startup
├── config.py                  # Environment variable validations & configuration objects
├── pyproject.toml / requirements.txt # Python dependencies
├── docker-compose.yml / Dockerfile   # Containerization config for the stack
├── app/
│   ├── database/              # Data models, DB initializations SQL (`create_tables.sql`)
│   ├── models/                # YOLO Weights (.pt) and TF Models (.h5, .pkl scalers)
│   └── services/              # Core ML processing APIs (RAG, Inference, Tracker)
├── data/                      # Raw Datasets & Chroma SQLite KB files
├── scripts/                   # Model Retraining Notebooks, retriever tests (`testing.py`)
├── static/                    # Frontend Assets (CSS, JS, Icons) + Image Uploads
└── templates/                 # Rendered UI Views (`index.html`, `analayze.html`, etc.)

---

Getting Started

Prerequisites

• Python: 3.11.x
• PostgreSQL: Version 14+ (Ensure a user is set up and permitted)
• Ollama: Installed and actively running locally (with the selected model pulled)

Local Development Setup (Windows PowerShell)

1. Clone & Virtual Environment

   python -m venv .venv
   .\.venv\Scripts\Activate.ps1

2. Install Dependencies

   python -m pip install --upgrade pip
   pip install -r requirements.txt

   If you do have a CUDA-compatible setup, use requirements_cuda.txt alternatives carefully and pin versions as needed for your machine.

3. Environment Variables

   Copy-Item .env.example .env

   Modify the new .env file to match your PostgreSQL credentials (DB_USER, DB_PASSWORD, DB_HOST, DB_NAME), the OPENWEATHERMAP_API_KEY, and ensure LLM_NAME / EMBEDDING_MODEL are aligned.

4. Pull Local LLM Model

   ollama pull llama3.1:8b

5. Initialize Database & RAG Check

   python bootstrap.py

6. Start Application Server

   python main.py

   Access the web app at http://localhost:5000

Docker Deployment

To spin up the entire application enclosed securely in its own environment with a managed Postgres container:

# Set DB_HOST=db in your .env before proceeding
docker compose up --build

On the very first launch, open a terminal into the container to run the bootstrap process:

docker compose exec app python bootstrap.py

---

Routes & User Flow

• Account Handling: Users sign up (/signup) or log in (/login) tracking device telemetry.
• Fields: Users manage their geographical plots of tea via /field/add specifying variety and age.
• Analyze Dashboard: Accessed via /analayze, farmers upload their images.

---

Background Jobs & Development Scripts

The /scripts/ folder houses research and ops protocols:

• train_model.py: Pipeline for updating the Keras recovery timeline model.
• report_generator.ipynb: Notebook analyzing historical dataset aggregates.
• retriever_correctness.py: Utility testing chroma extraction vs. expected ground truths for debugging.