🌱 AgroCare — Plant Disease Detection (Streamlit + MongoDB + LLM Agents)

A multi-version Streamlit app (V1, V2, V3) for plant disease detection, comparison, and history tracking, backed by MongoDB and powered by multi‑agent LLM analysis. Includes secure auth (bcrypt) and data persistence for repeatable diagnostics.

---

▶️ Demo

---

✨ Features

• Three app versions kept for transparency & evolution
  • V1 — Plant Disease Detection: Upload a plant image → get symptoms, likely diseases, and care steps.
  • V2 — Image Comparison: Compare two images (e.g., before vs. after treatment) → track improvement or spread; results are persisted for future comparisons.
  • V3 — Scanned Plants & Details: History view of analyzed plants, detail pages, follow‑up image attachments, and side‑by‑side analysis.
• Secure Auth using bcrypt:
  • Passwords are salted + hashed, verified using constant‑time checks.
  • Login/Signup state managed via cookies; only minimal identifiers stored client‑side.
  • User records persisted in MongoDB.
• Multi‑Agent Analysis for higher accuracy:
  • Agents collaborate: Vision Analyzer → Symptom Summarizer → Differential Diagnosis → Consistency Checker.
  • Modular provider layer (supports multiple APIs) to ensemble or fallback.
• MongoDB Persistence:
  • Image binaries, analyses, and comparisons saved per‑user (owner scoping).
  • Queryable history enables longitudinal tracking and re‑evaluation.
• Streamlit UX with caching:
  • st.cache_resource for long‑lived resources (DB clients, model clients).
  • st.cache_data for deterministic, recomputable results (lookups, small transforms).

---

🧰 Tech Stack

---

🔐 Security

• Passwords: hashed with bcrypt (salted), verified with bcrypt.checkpw.
• Cookies/Sessions: store only minimal identifiers (e.g., uid), prefer HttpOnly, Secure, SameSite=Lax/Strict.
• Input validation: accept only JPEG/PNG, enforce size limits, verify magic bytes.
• Secrets: never commit real keys; use config/secrets.example.toml template and local .streamlit/secrets.toml.
• Least privilege: use a scoped MongoDB user (RBAC) and IP allow‑listing where possible.

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• MongoDB Atlas (or local Mongo) connection string

Clone & Install

git clone https://github.com/aymenhmaidiwastaken/AgroCare.git
cd AgroCare
python -m venv .venv && source .venv/bin/activate   # Windows: .venv\\Scripts\\activate
pip install -r requirements.txt

Configure Secrets (template → local)

# copy the example template
mkdir -p .streamlit
cp config/secrets.example.toml .streamlit/secrets.toml

# then edit .streamlit/secrets.toml with your values
[mongodb]
uri = "your-mongodb-uri"

[api_keys]
openai = "sk-..."
# optional providers, if configured
# groq = "..."
# toolhouse = "..."

Run

streamlit run app.py

Use the sidebar to switch between V1 / V2 / V3.

---

🧪 Multi‑Agent Pipeline (high level)

1. Vision Analyzer extracts salient features from the image (spots, color changes, leaf curl).
2. Symptom Summarizer condenses features into a structured list.
3. Differential Diagnosis proposes 1–2 likely diseases with confidence levels.
4. Consistency Checker verifies plausibility and flags contradictions.
5. Persistence writes final analysis/trace to MongoDB for future review.

Each step can target different APIs/models and ensemble results for robustness.

---

📦 Project Layout

AgroCareApp/
├─ app.py                      # main Streamlit entry (routes to V1/V2/V3)
├─ V1/                         # Plant Disease Detection
│  └─ app1.py
├─ V2/                         # Image Comparison
│  └─ app2.py
├─ V3/                         # Scanned Plants / Details
│  └─ app3.py
├─ assets/
│  ├─ demo.gif                 # your showcase GIF
│  └─ icons/                   # tech logos (svg/png)
├─ config/
│  └─ secrets.example.toml     
├─ requirements.txt
├─ .gitignore
└─ README.md

---

🗺️ Roadmap

• Role‑based access (admin/user)
• GridFS for large images
• Structured JSON outputs for analyses (Pydantic)
• Model confidence calibration & uncertainty display
• Batch comparisons and alerts ("notify me if worsening")

---

🤝 Contributing

Pull requests welcome. If you find a bug or security issue, please open an issue or email us (see SECURITY.md for responsible disclosure).

---