commit 5e24d401bc13a699fdcb5f3a80e0ed016d803846
parent 724604d76ee3783f075310c7aeced5c2b093d51d
Author: maydayv7 <maydayv7@gmail.com>
Date: Thu, 26 Mar 2026 19:35:30 +0530
Update README
Diffstat:
10 files changed, 324 insertions(+), 390 deletions(-)
diff --git a/agent/Qdrant/Search.py b/agent/Qdrant/Search.py
diff --git a/assets/screenshots/agent_control.png b/assets/screenshots/agent_control.png
Binary files differ.
diff --git a/assets/screenshots/demeter_dashboard.png b/assets/screenshots/demeter_dashboard.png
Binary files differ.
diff --git a/assets/screenshots/landing_page.png b/assets/screenshots/landing_page.png
Binary files differ.
diff --git a/assets/screenshots/system_overview.png b/assets/screenshots/system_overview.png
Binary files differ.
diff --git a/backend/server/functions.py b/backend/server/functions.py
@@ -751,7 +751,7 @@ async def process_similar_crops(crop_id: str, crop_name: str, payload_json: str)
]
)
- search_results = client.search(
+ search_results = client.query_points(
collection_name=COLLECTION_NAME,
query_vector=query_vector,
query_filter=exclude_filter,
diff --git a/frontend/src/utils/translations.js b/frontend/src/utils/translations.js
@@ -12,7 +12,7 @@ const en = {
nav_crops_need_attention: "{n} crop{s} need attention",
nav_all_clear: "All clear",
nav_harvest_ready: "๐พ {n} ready to harvest",
- sidebar_agri_ai: "AGRIยทAIยทv2",
+ sidebar_agri_ai: "AGRIยทAIยทv4",
sidebar_connecting: "CONNECTINGโฆ",
sidebar_farm_online: "FARM ONLINE",
sidebar_no_data: "NO DATA",
diff --git a/readme.md b/readme.md
@@ -1,229 +1,329 @@
-# Demeter: Autonomous Hydroponic Intelligence ๐ฟ๐ค
+# Demeter - Autonomous Hydroponic Intelligence
-<a href="https://drive.google.com/file/d/1VAN31mXPaQ7r4Fm8dpzjhgGeeQwvlH-Z/view?usp=drive_link">
- <img src="https://img.shields.io/badge/Demeter-Hydroponic_AI-4CAF50?style=for-the-badge&logo=robot&logoColor=white" alt="Demeter Logo">
-</a>
+> _"The farm thinks for itself."_
-<div align="center">
+Demeter is a fully autonomous, multi-agent AI system for precision hydroponic farm management.
+It combines reinforcement learning, computer vision, LLM-based reasoning, vector memory, and a physics-based digital twin to monitor, analyze, and act on crop conditions - **24/7, without human intervention.**
-**Industrial-grade Multi-Agent System for autonomous hydroponic farming through AI-driven reasoning**
+Built for the **Microsoft AI Unlocked - AI for India** hackathon, Demeter addresses food security and precision agriculture challenges in the Indian context, with full Hindi language support and a design philosophy accessible to rural operators.
-[๐ Documentation](https://drive.google.com/file/d/1VAN31mXPaQ7r4Fm8dpzjhgGeeQwvlH-Z/view?usp=drive_link) โข [๐ Quick Start](#-quick-start) โข [๐ง API Reference](#-api-reference)
+---
+
+## Screenshots
-</div>
+
-<div align="center">
+---
-|  |  |  |
-| :--------------------------------------------------------: | :----------------------------------------------------: | :------------------------------------------------: |
-| **System Overview**<br/>Real-time monitoring dashboard | **Agent Control**<br/>Multi-agent orchestration | **Console Log**<br/>AI agent decision logs |
+## Key Features
-</div>
+### Multi-Agent System (MAS)
----
+- **7 specialized AI agents** collaborating in a structured pipeline
+- **Contextual Bandit (MARL)** for cycle-level strategic goal selection based on historical rewards
+- **LangGraph state machines** for each expert agent with tool-use and retry loops
+- **Supervisor Agent** that synthesizes conflicting expert plans and enforces safety constraints
-## ๐ Overview
+### AI & Machine Learning
-**Demeter** is a revolutionary Multi-Agent System (MAS) that transforms hydroponic farming through intelligent automation. Unlike traditional rule-based systems that react to thresholds, Demeter employs a cognitive architecture that **perceives, reasons, and acts** like an expert grower.
+- **Azure OpenAI (GPT-4.1)** powering all LLM-based reasoning agents
+- **Azure Custom Vision** for real-time plant disease and pest detection
+- **CLIP (ViT-B/32)** for plant image encoding into the FMU vector
+- **Custom Contextual Bandit** (LinGreedy) for strategic action selection
+- **RAG pipeline** with FastEmbed + Qdrant for domain knowledge retrieval
-The system combines **Long-Term Memory**, **Computer Vision**, and **Reinforcement Learning** to optimize crop health in real-time, creating a truly autonomous farming experience.
+### Memory & Storage
-<div align="center">
- <img src="assets/screenshots/demeter_dashboard.png" alt="Demeter Dashboard" width="800"/>
- <p><em>Demeter Web Dashboard - Real-time monitoring and control interface</em></p>
-</div>
+- **Qdrant Vector DB** - dual-collection architecture:
+ - `Crop_States` - FMU snapshots (image + sensor fused vectors)
+ - `Knowledge_Base` - agronomic research documents (RAG)
+ - `Plant_Biographies_HF` - long-term per-crop memory (via Mem0)
+- **Mem0** - semantic plant biography system backed by Azure OpenAI
+- **Node.js + MongoDB** - structured crop metadata and event logs
-### ๐ฏ Key Capabilities
+### Physics Simulator
-- **๐ง Cognitive Decision Making**: AI agents that reason like human experts
-- **๐ Real-time Disease Detection**: Azure Custom Vision-powered visual diagnosis
-- **๐ Scientific Knowledge Base**: RAG-enabled agricultural research integration
-- **๐ฎ Adaptive Learning**: Reinforcement learning that improves over time
-- **๐ Live Data Integration**: Autonomous web search for current conditions
-- **โก Predictive Modeling**: Digital twin simulation before actions
-- **๐ Safety-First Design**: Multi-layer validation prevents harmful actions
+- **Digital Twin** with a hybrid physics + neural residual model
+- Simulates pH, EC, water temp, air temp, humidity, VPD, and biomass
+- Exposes REST API consumed by both the agent loop and frontend
+- Syncs state to **Azure Digital Twins** after every action
---
-## ๐๏ธ System Architecture
+## Tech Stack
+
+| Layer | Technology |
+| ----------------------- | ---------------------------------------------------- |
+| **LLM** | Azure OpenAI GPT-4.1 |
+| **Agent Orchestration** | LangGraph, LangChain |
+| **Vision** | Azure Custom Vision, CLIP ViT-B/32 (OpenAI) |
+| **Vector DB** | Qdrant |
+| **Semantic Memory** | Mem0 |
+| **Embeddings** | FastEmbed (BAAI/bge-small-en-v1.5), all-MiniLM-L6-v2 |
+| **Digital Twin** | Azure Digital Twins + custom physics sim |
+| **Backend (Python)** | FastAPI, Uvicorn |
+| **Backend (Node)** | Express.js, MongoDB |
+| **Frontend** | React 19, React Router v7, Recharts, Tailwind CSS v3 |
+| **Physics** | PyTorch (ResidualPhysicsNet), NumPy |
+| **RL** | Custom Contextual Bandit (LinGreedy) |
-Demeter operates on a **Hierarchical Control Loop** powered by **LangGraph**, featuring specialized agents that collaborate to maintain optimal growing conditions:
+---
-<div align="center">
- <img src="assets/architecture.png" alt="Demeter System Architecture" width="800"/>
- <p><em>Demeter Agent Hierarchy - Multi-agent cognitive architecture</em></p>
-</div>
+## Architecture
-### ๐ค Agent Roles
+
-| Agent | Role | Technology | Purpose |
-| --------------- | ------------- | -------------------- | --------------------------------------------- |
-| **Supervisor** | Executive | Contextual Bandit RL | Strategic decision making & safety validation |
-| **Researcher** | Scholar | RAG + Web Search | Scientific consultation & live data retrieval |
-| **Judge** | Auditor | CV + Analytics | Performance evaluation & RL training |
-| **Atmospheric** | Specialist | Physics Engine | VPD, CO2, light optimization |
-| **Water** | Specialist | Chemistry Engine | pH, EC, nutrient balancing |
-| **Doctor** | Diagnostician | Azure Custom Vision | Disease detection & visual analysis |
-| **Historian** | Memory | Mem0 + Qdrant | Long-term plant biography & context |
+### The Farm Management Unit (FMU)
----
+The FMU is the core data structure of Demeter.
+It is a **fused multimodal vector** created by the Sentinel system at the start of every cycle:
-## โจ Key Features
+```
+Image (512ร512 plant photo)
+ โ CLIP ViT-B/32
+ [512-dim visual embedding]
+ +
+Sensor Window (pH, EC, temp, humidity, VPD, biomass, ...)
+ โ SensorEncoder (LSTM + linear projection)
+ [7-dim sensor embedding]
+ =
+FMU Vector [519-dim] โโ stored in Qdrant with full metadata payload
+```
-### โก Self-Correcting Reasoning
+The FMU is stored in Qdrant and becomes the **memory of the farm** - every past state is retrievable by similarity search.
-- **Digital Twin Simulation**: Predicts action consequences before execution, powered by Azure Digital Twin
-- **Safety Interlocks**: Prevents harmful actions through multi-layer validation
-- **Rollback Capabilities**: Can reverse unsafe decisions
+---
-### ๐ RAG-Powered Knowledge Base
+### The 7-Agent Pipeline
-- **Scientific Literature**: Indexes agricultural research papers and best practices
-- **Contextual Retrieval**: Retrieves relevant information for current conditions
-- **Hallucination Prevention**: All decisions grounded in verified sources
+Each autonomous cycle follows this sequence:
-### ๐ฏ Reinforcement Learning Optimization
+```
+โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
+โ DEMETER AGENT CYCLE โ
+โ โ
+โ 1. FetchingAgent โ
+โ โโ Polls simulator/ADT โ builds FMU โ stores in Qdrant โ
+โ โ
+โ 2. JudgeAgent (LangGraph) โ
+โ โโ Retrieves N-1 FMU โ runs Azure CV (visual forensics) โ
+โ โโ Queries FarmMemory (Mem0) โ deliberates via LLM โ
+โ โโ Files reward score โ trains Bandit โ
+โ โ
+โ 3. SupervisorAgent โ ContextualBandit โ
+โ โโ Encodes current FMU as context vector (519-dim) โ
+โ โโ Bandit selects 1 of 15 strategies (LinGreedy) โ
+โ โโ Strategy instruction passed to expert agents โ
+โ โ
+โ 4. ResearcherAgent โ
+โ โโ Queries Knowledge_Base (RAG) for crop-specific data โ
+โ โ
+โ 5. AtmosphericAgent (LangGraph) โ
+โ โโ Tools: ask_rag, web_search, calculate_vpd, diagnose_plantโ
+โ โโ Generates climate action plan (temp/humidity/CO2/light) โ
+โ โโ Simulates plan โ retry loop if unsafe โ
+โ โ
+โ 6. WaterAgent (LangGraph) โ
+โ โโ Tools: ask_rag, web_search, calculate_vpd, diagnose_plantโ
+โ โโ Generates nutrient plan (pH dosing, EC adjustment) โ
+โ โโ Simulates plan โ retry loop if unsafe โ
+โ โ
+โ 7. SupervisorAgent โ synthesize_plan โ
+โ โโ Merges atmospheric + water plans โ
+โ โโ Detects conflicts, enforces safety bounds โ
+โ โโ Dispatches final FarmAction to Simulator/ADT โ
+โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
+```
-- **Contextual Bandit Algorithm**: Learns optimal strategies over time
-- **Adaptive Decision Making**: Improves performance based on outcomes
-- **Strategy Evolution**: Discovers better approaches through trial and feedback
+### The 15-Strategy MARL Bandit
-### ๐๏ธ Advanced Computer Vision
+The Contextual Bandit (`agent/Marl/bandit.py`) is a **LinGreedy** (pure-exploitation) contextual bandit that selects from 15 farm management strategies:
-- **Real-time Disease Detection**: Identifies pathogens before symptoms appear
-- **Growth Stage Analysis**: Monitors plant development and health indicators
-- **Automated Documentation**: Creates visual records of plant conditions
+| Category | Strategies |
+| ------------ | ----------------------------------------------------------------------------------------------- |
+| ๐ข Passive | MAINTAIN_CURRENT, CALIBRATE_SENSORS |
+| ๐ง Nutrients | AGGRESSIVE_PH_DOWN/UP, GENTLE_PH_BALANCING, INCREASE_EC_VEG/BLOOM, LOWER_EC_FLUSH, CALMAG_BOOST |
+| ๐ค๏ธ Climate | RAISE_TEMP_HUMIDITY, LOWER_TEMP_HUMIDITY, MAX_AIR_CIRCULATION |
+| ๐ Disease | FUNGAL_TREATMENT, PEST_ISOLATION, PRUNE_NECROTIC_LEAVES |
-### ๐ Autonomous Intelligence
+The bandit uses the **519-dim FMU vector as its context**, learns reward signals from the JudgeAgent, and persists its model weights to disk (`model_bandit_greedy.pkl`).
-- **Live Web Search**: Fetches current weather, market data, and research
-- **Dynamic Knowledge Updates**: Integrates new information without redeployment
-- **Environmental Adaptation**: Adjusts to local conditions and climate changes
+### The JudgeAgent - Closing the RL Loop
----
+The JudgeAgent is the reward signal generator. After each cycle, it:
-## ๐ ๏ธ Technology Stack
+1. Retrieves the previous FMU (N-1) from Qdrant
+2. Runs **Azure CV-based visual diagnosis** on the plant image
+3. Queries **Mem0** for the crop's text biography
+4. **Deliberates** via LLM: compares sensor delta, visual report, and history
+5. Issues a reward score (โ1.0 to +1.0) and outcome classification
+6. **Updates Qdrant** payload and **writes to FarmMemory (Mem0)**
+7. Returns training data โ Bandit updates its weights online
-### Backend (AI Brain)
+### The Simulator - Physics + Neural Residual Digital Twin
-```python
-# Core Framework
-- FastAPI 0.109+ # High-performance async API
-- Python 3.11+ # Modern Python with performance optimizations
+The simulator (`simulator/main.py`) is a **FastAPI server** running a `DigitalTwin` class:
-# AI Orchestration
-- LangChain 0.1+ # LLM orchestration framework
-- LangGraph 0.0.26+ # Multi-agent workflow management
+- **Hybrid physics model**: First-principles equations for pH, EC, VPD, biomass growth
+- **Neural residual**: A small `ResidualPhysicsNet` (PyTorch MLP) that corrects physics approximations
+- **State vector**: `[pH, EC, water_temp, air_temp, humidity, VPD, biomass]`
+- **Image generation**: Outputs plant health images based on a bucket score (0โ100)
+- **Azure Digital Twins sync**: Every action call pushes telemetry to the ADT twin (`HydrophonicTank`)
+- **Dual endpoints**: `/simulation/state` (agent loop) and `/azure/state` (ADT read-back)
-# AI Models
-- Azure Custom Vision # Object detection for disease identification
-- Llama-3.3-70b (Groq) # Primary LLM for reasoning
-- OpenAI GPT-4o # Fallback LLM option
-- CLIP (OpenAI) # Vision-language understanding
+---
-# Data & Memory
-- Azure Digital Twin # Digital farm simulation
-- Qdrant # Vector database for RAG and embeddings
-- Mem0 # Semantic long-term memory
-- FastEmbed # Local embedding generation
-```
+## Frontend
-### Frontend (User Interface)
+The frontend is a highly responsive React 19 SPA featuring bilingual support (English/เคนเคฟเคจเฅเคฆเฅ) for accessibility in Indian agriculture, made to serve as a Proof-of-Concept.
-```javascript
-- React 19+ # Modern UI framework
-- React Router 7+ # Client-side routing
-- Tailwind CSS 3+ # Utility-first styling
-- Recharts 3+ # Data visualization
-- Lucide React # Icon library
-```
+### Pages
-### Infrastructure
+| Page | Route | Description |
+| ----------------- | --------------- | -------------------------------------------------------------------------------------------------- |
+| Landing Page | `/` | Hero, live agent activity feed, feature cards, live stats |
+| Dashboard | `/dashboard` | Crop card grid with health/maturity/stage filters and harvest banner |
+| Crop Details | `/crop/:id` | Per-crop sensor charts, agent reasoning log, event timeline, actuator commands |
+| Add Crop | `/add-crop` | Form to initialize a crop + live agent pipeline log with 6-phase progress tracker |
+| Farm Intelligence | `/intelligence` | Natural-language RAG search against the agronomic knowledge base |
+| Analytics | `/analytics` | Multi-chart analytics: pH/EC/temp traces, daily sequences, parameter health scores, per-crop table |
+| Alerts | `/alerts` | Categorized alert system (CRITICAL/WARNING/INFO/HARVEST) with acknowledge workflow |
+| Settings | `/settings` | Theme, language, farm name, notification prefs, onboarding |
-- **Database**: Qdrant (Vector Search)
-- **Deployment**: Docker containers
-- **Monitoring**: Built-in logging and health checks
-- **Security**: API key authentication and validation
+---
+
+## Project Structure
+
+```
+demeter/
+โโโ agent/
+โ โโโ main_agent.py # Orchestration loop (runs the 7-agent cycle)
+โ โโโ memory.py # FarmMemory class (Mem0 + Qdrant)
+โ โโโ Marl/
+โ โ โโโ bandit.py # Contextual Bandit (LinGreedy, 15 arms)
+โ โ โโโ strategies.py # Strategy definitions
+โ โ โโโ train-bandit.py # Offline training script
+โ โโโ Sentinel/
+โ โ โโโ agent.py # FMUBuilder - creates fused vectors
+โ โ โโโ fmu.py # FMU dataclass
+โ โ โโโ Encoders/
+โ โ โโโ Vision.py # CLIP ViT-B/32 image encoder
+โ โ โโโ TimeSeries.py # LSTM-based sensor encoder
+โ โโโ Qdrant/
+โ โ โโโ Client.py # Qdrant client singleton
+โ โ โโโ Setup.py # Collection initialization
+โ โ โโโ Store.py # FMU storage helpers
+โ โ โโโ Search.py # Similarity search helpers
+โ โโโ sub_agents/
+โ โ โโโ base_agent.py # BaseReasoningAgent (Azure OpenAI client)
+โ โ โโโ fetching_agent.py # FetchingAgent - polls simulator, builds FMU
+โ โ โโโ judge_agent.py # JudgeAgent - reward evaluation (LangGraph)
+โ โ โโโ atmospheric_agent.py # AtmosphericAgent - climate planning (LangGraph)
+โ โ โโโ water_agent.py # WaterAgent - nutrient planning (LangGraph)
+โ โ โโโ Supervisor.py # SupervisorAgent - strategy + synthesis
+โ โ โโโ Researcher.py # ResearcherAgent - RAG knowledge retrieval
+โ โ โโโ Explainer.py # ExplainerAgent - chain-of-thought log generation
+โ โ โโโ Doctor.py # VisionAgent - plant disease detection (Azure Custom Vision)
+โ โ โโโ water_and_atmospheric_dependencies/
+โ โ โโโ state.py # LangGraph AgentState definition
+โ โ โโโ nodes.py # LangGraph node functions
+โ โ โโโ tools.py # LangChain tools (calculate_vpd, web_search)
+โ โ โโโ physics_engine.py# LLM-based plan safety simulator
+โ โ โโโ retrieval.py # LangChain tools (ask_rag, diagnose_plant, ask_memory)
+โ โโโ tools/
+โ โโโ actuation.py # Actuator command builders
+โ โโโ db_tools.py # Database utility tools
+โ โโโ processing_tools.py # Sensor processing utilities
+โ โโโ reset_memory.py # Memory reset utility
+โ
+โโโ backend/
+โ โโโ server/
+โ โ โโโ main.py # FastAPI server (Python) - agent HTTP endpoints
+โ โ โโโ functions.py # Backend utility functions
+โ โ โโโ rag_brain.py # PDF ingestion pipeline for Knowledge Base
+โ โ โโโ create-index.py # Qdrant index setup script
+โ โ โโโ reset-db.py # Database reset utility
+โ โโโ node_server/
+โ โโโ index.js # Express.js server - crop CRUD API
+โ โโโ routes/farmRoutes.js # Farm route definitions
+โ โโโ controllers/ # Farm controller logic
+โ โโโ config/db.js # MongoDB connection
+โ
+โโโ frontend/
+โ โโโ src/
+โ โ โโโ App.js # Router, providers, onboarding gate
+โ โ โโโ pages/ # All 8 page components
+โ โ โโโ components/ # Sidebar, AgentWidgets, Onboarding
+โ โ โโโ hooks/ # useFarmData, useSettings, useTranslation
+โ โ โโโ api/ # agentApi.js, farmApi.jsx
+โ โ โโโ utils/ # translations.js, dataUtils.js
+โ โ โโโ data/mockData.js # Mock data for testing
+โ โโโ tailwind.config.js
+โ
+โโโ simulator/
+โ โโโ main.py # DigitalTwin + FastAPI server + Azure ADT sync
+โ
+โโโ Knowledge_Base/ # Drop agronomic PDFs here for RAG ingestion
+โโโ requirements.txt
+โโโ README.md
+```
---
-## ๐ Quick Start
+## Setup & Installation
### Prerequisites
-Before installing Demeter, ensure you have:
-
-- **Python 3.10+** - [Download](https://www.python.org/downloads/)
-- **Node.js 16+** - [Download](https://nodejs.org/)
-- **Git** - [Download](https://git-scm.com/)
-- **Docker Desktop** (for local Qdrant) OR [Qdrant Cloud account](https://cloud.qdrant.io/)
+- Python 3.10+
+- Node.js 18+
+- A running [Qdrant](https://qdrant.tech/) instance (local Docker or Qdrant Cloud)
+- Azure OpenAI resource with GPT-4.1 deployment
+- (Optional) Azure Digital Twins instance
-### 1. Clone and Setup
+### 1. Clone & Install Python Dependencies
```bash
-# Clone the repository
git clone https://github.com/maydayv7/demeter.git
cd demeter
-
-# Create virtual environment
-python -m venv .venv
-source .venv/bin/activate # Windows: .venv\Scripts\activate
-
-# Install Core Python dependencies
-pip install -r requirements.txt
-
-# Install Simulator dependencies
-cd simulator
pip install -r requirements.txt
-cd ..
-
-# Install Backend Node dependencies
-cd backend/node_server
-npm install
-cd ../..
-
-# Install Frontend dependencies
-cd frontend
-npm install
-cd ..
```
-### 2. Environment Configuration
+### 2. Configure Environment Variables
-Create a **single** `.env` file in the **project root directory**. All components are configured to read from this top-level file automatically.
+Create a `.env` file in the project root:
```env
-# Database
-QDRANT_URL=http://localhost:6333
-QDRANT_API_KEY=your_qdrant_key_here
+# Azure OpenAI
+AZURE_OPENAI_API_KEY=your_azure_openai_key
+AZURE_OPENAI_ENDPOINT=https://your-resource.openai.azure.com/
+AZURE_OPENAI_DEPLOYMENT_NAME=gpt-4.1
+AZURE_OPENAI_API_VERSION=2024-12-01-preview
-# LLM
-GROQ_API_KEY=gsk_your_key_here
+# Qdrant
+QDRANT_URL=http://localhost:6333
+QDRANT_API_KEY=your_qdrant_api_key # Only needed for Qdrant Cloud
+
+# Azure Custom Vision
+AZURE_API_KEY=your_azure_api_key
+AZURE_ENDPOINT=your_azure_endpoint
+AZURE_PROJECT_ID=your_azure_project_id
+AZURE_PREDICTION_KEY=your_azure_prediction_key
+AZURE_URL=your_azure_url
+AZURE_ITERATION_NAME=DemeterDoctor
+DATASET_FOLDER="train"
# Simulator
-SIMULATOR_PORT=8001
-SIMULATOR_ACTION_URL=http://localhost:8001/simulation/action
SIMULATOR_STATE_URL=http://localhost:8001/simulation/state
-ADT_URL=simulator.api.krc.digitaltwins.azure.net
-AZURE_TENANT_ID=your_azure_tenant_id_here
-AZURE_CLIENT_ID=your_service_principal_client_id_here
-AZURE_CLIENT_SECRET=your_service_principal_client_secret_here
+SIMULATOR_ACTION_URL=http://localhost:8001/simulation/action
+SIMULATOR_PORT=8001
-# Node Backend
-PORT=3001
+# Azure Digital Twins
+ADT_URL=your-adt-instance.digitaltwins.azure.net
-# React Frontend
+# Frontend
REACT_APP_AGENT_API_URL=http://localhost:8000
REACT_APP_FARM_API_URL=http://localhost:3001/api
-
-# Azure Services
-AZURE_API_KEY=your_azure_key_here
-AZURE_ENDPOINT=azure_endpoint_here
-AZURE_PROJECT_ID=your_azure_project_id_here
-DATASET_FOLDER="train"
-AZURE_PREDICTION_KEY=your_azure_predict_key_here
-AZURE_URL=your_azure_url_here
-AZURE_ITERATION_NAME=DemeterDoctor
+PORT=3001
```
### 3. Start Qdrant Database
@@ -247,220 +347,53 @@ Run the following command from the root directory to create the required collect
python backend/server/create-index.py
```
-### 5. Start the Simulator
+### 5. Initialize the Knowledge Base (RAG)
-Open a new terminal, activate the virtual environment, and run the Digital Twin Simulator from its directory:
+Place agronomic PDFs in the `Knowledge_Base/` folder, then run:
```bash
-cd simulator
-python main.py
+python backend/server/rag_brain.py
```
-### 6. Start the Agents
-
-Download [model_bandit_greedy.pkl](https://drive.google.com/file/d/1spuw3TogZRtP0fYZkYA2Kxz1-CiUzBDA/view?usp=drive_link) and [plant_disease_model.pt](https://drive.google.com/file/d/1NkdGt0CFS7tx4vttp8Tod8ksjDLib8dp/view?usp=drive_link) and place them under `agent/Marl` and `agent/model` respectively.
+Otherwise, download [model_bandit_greedy.pkl](https://drive.google.com/file/d/1spuw3TogZRtP0fYZkYA2Kxz1-CiUzBDA/view?usp=drive_link) and place it inside the `agent/Marl` directory.
-Open a new terminal, activate the virtual environment, and start the agent orchestrator from the agent directory:
+### 6. Start the Simulator
```bash
-cd agent
-python main_agent.py
+cd simulator
+python main.py
+# Runs on http://localhost:8001
```
-### 7. Start the Backends
+### 7. Start the Backend Servers
-**Python API Server:**
-Open a new terminal, activate the virtual environment, and start the FastAPI server from the backend directory:
+Download [plant_disease_model.pt](https://drive.google.com/file/d/1NkdGt0CFS7tx4vttp8Tod8ksjDLib8dp/view?usp=drive_link) and place it inside the `agent/model` directory.
```bash
+# Python FastAPI server
cd backend/server
-python main.py
-```
+uvicorn main:app --reload --port 8000
-**Node.js Database Server:**
-Open a new terminal and start the Express server for fetching from the memory database:
-
-```bash
+# Node.js Express server
cd backend/node_server
-npm start
+npm install && npm start
```
-### 8. Run the Frontend
-
-Open a new terminal, navigate to the frontend directory, and run the development server:
+### 8. Start the Frontend
```bash
cd frontend
+npm install
npm start
+# Runs on http://localhost:3000
```
-### 9. Access the Application
-
-- **Frontend**: http://localhost:3000
-- **API Documentation**: http://localhost:8000/docs
-- **Health Check**: http://localhost:8000/health
-
-<div align="center">
- <img src="assets/screenshots/agent_control.png" alt="Agent Control Interface" width="700"/>
- <p><em>Agent Control Dashboard - Monitor and interact with AI agents in real-time</em></p>
-</div>
-
----
-
-## ๐ Project Structure
-
-```
-demeter/
-โโโ agent/ # AI Agent System
-โ โโโ main_agent.py # Main orchestrator
-โ โโโ sub_agents/ # Specialized agents
-โ โ โโโ Supervisor.py # Executive decision maker
-โ โ โโโ Researcher.py # RAG-powered research
-โ โ โโโ judge_agent.py # Performance auditor
-โ โ โโโ atmospheric_agent.py # Climate control
-โ โ โโโ water_agent.py # Hydroponics management
-โ โ โโโ Doctor.py # Disease diagnostician
-โ โโโ Qdrant/ # Vector database clients
-โ โโโ tools/ # Agent utilities
-โ โโโ memory.py # Long-term memory system
-โ โโโ model/ # ML models
-โโโ backend/ # API Server
-โ โโโ server/
-โ โ โโโ main.py # FastAPI application
-โ โ โโโ functions.py # Business logic
-โ โ โโโ rag_brain.py # AI integration
-โ โโโ node_server/ # Additional API endpoints
-โโโ frontend/ # React Application
-โ โโโ src/
-โ โ โโโ components/ # UI components
-โ โ โโโ pages/ # Application pages
-โ โ โโโ api/ # API integration
-โ โโโ public/ # Static assets
-โโโ requirements.txt # Python dependencies
-โโโ README.md # This file
-```
-
----
-
-## ๐ง API Reference
-
-### Core Endpoints
-
-| Method | Endpoint | Description |
-| ------ | ------------------------ | -------------------- |
-| `GET` | `/health` | System health check |
-| `GET` | `/api/farms` | List all farms |
-| `POST` | `/api/farms` | Create new farm |
-| `GET` | `/api/farms/{id}` | Get farm details |
-| `POST` | `/api/agents/action` | Trigger agent action |
-| `GET` | `/api/memory/{plant_id}` | Get plant history |
-
-### Agent Control
-
-```bash
-# Get current system status
-curl http://localhost:8000/health
-
-# Trigger manual agent cycle
-curl -X POST http://localhost:8000/api/agents/action \
- -H "Content-Type: application/json" \
- -d '{"action": "analyze", "farm_id": "farm_001"}'
-
-# Query plant memory
-curl http://localhost:8000/api/memory/plant_123
-```
-
-### WebSocket Real-time Updates
-
-```javascript
-// Connect to real-time updates
-const ws = new WebSocket("ws://localhost:8000/ws/farm-updates");
-
-ws.onmessage = (event) => {
- const data = JSON.parse(event.data);
- console.log("Farm update:", data);
-};
-```
-
----
-
-## ๐ Monitoring & Troubleshooting
-
-### Health Checks
+### 9. Run the Agent Loop
```bash
-# Check system health
-curl http://localhost:8000/health
-
-# Check agent status
-curl http://localhost:8000/api/agents/status
-
-# View logs
-tail -f logs/demeter.log
+python agent/main_agent.py
```
-### Common Issues
-
-**Q: Agents not responding**
-
-- Check Qdrant connection: `curl http://localhost:6333/health`
-- Verify API keys in `.env`
-- Ensure virtual environment is activated
-
-**Q: Memory not persisting**
-
-- Check Qdrant collections: Access Qdrant dashboard
-- Verify embedding model is loaded
-- Check disk space and permissions
-
-**Q: Vision analysis failing**
-
-- Check camera/image permissions
-- Verify OpenCV installation
-
-**Q: Web search not working**
-
-- Validate SerpAPI key
-- Check internet connectivity
-- Review API quota limits
-
-### Performance Tuning
-
-```python
-# Adjust agent cycle frequency
-AGENT_CYCLE_INTERVAL = 300 # seconds
-
-# Configure memory limits
-MAX_MEMORY_ENTRIES = 10000
-
-# Set vision model confidence threshold
-VISION_CONFIDENCE_THRESHOLD = 0.7
-```
-
----
-
-## ๐ Performance Metrics
-
-### System Benchmarks
-
-- **Decision Latency**: <2 seconds per cycle
-- **Memory Retrieval**: <500ms average
-- **Vision Analysis**: <1 second per image
-- **RAG Query**: <300ms average
-- **Uptime**: 99.9% target
-
-### Accuracy Metrics
-
-- **Disease Detection**: 94% accuracy (Azure CV fine-tuned)
-- **Decision Quality**: 89% optimal actions (RL trained)
-- **Safety Compliance**: 100% (validation enforced)
-
---
-<div align="center">
-
**Made with โค๏ธ for the future of sustainable agriculture**
-
-[๐ Star us on GitHub](https://github.com/maydayv7/demeter) โข [๐ Report a bug](https://github.com/maydayv7/demeter/issues) โข [๐ก Request a feature](https://github.com/maydayv7/demeter/issues/new?template=feature_request.md)
-
-</div>
diff --git a/requirements.txt b/requirements.txt
@@ -1,39 +1,50 @@
-# --- Core Framework & Server ---
-fastapi>=0.109.0
-uvicorn[standard]>=0.27.0
-python-multipart>=0.0.9
-python-dotenv>=1.0.0
-pydantic>=2.6.0
-requests>=2.31.0
-
-# --- AI Orchestration (LangChain) ---
-langchain>=0.1.0
-langchain-core>=0.1.10
-langchain-openai>=0.0.5
-langgraph>=0.0.26
-langchain-community>=0.0.10
-
-# --- Database & Memory ---
-qdrant-client>=1.7.0
-mem0ai>=0.0.12
-fastembed>=0.2.0
-
-# --- Computer Vision & Machine Learning ---
+# Core Framework & Server
+fastapi>=0.115.0
+uvicorn[standard]>=0.30.0
+python-multipart>=0.0.12
+python-dotenv>=1.0.1
+pydantic>=2.9.0
+requests>=2.32.0
+httpx>=0.27.0
+aiofiles>=24.1.0
+
+# AI Orchestration
+langchain>=0.3.0
+langchain-core>=0.3.0
+langchain-openai>=0.2.0
+langgraph>=0.2.0
+langchain-community>=0.3.0
+
+# Azure SDKs
+openai>=1.50.0
+azure-identity>=1.19.0
+azure-digitaltwins-core>=1.2.0
+
+# Vector Database & Memory
+qdrant-client>=1.12.0
+mem0ai>=0.1.0
+fastembed>=0.4.0
+
+# Computer Vision & Machine Learning
numpy>=1.26.0
-torch>=2.2.0
-torchvision>=0.17.0
-opencv-python>=4.9.0.80
-Pillow>=10.2.0
-
-# --- CLIP (OpenAI) ---
-# Required for Sentinel/Vision Encoders
-ftfy
-regex
-tqdm
+torch>=2.3.0
+torchvision>=0.18.0
+opencv-python>=4.10.0
+Pillow>=10.4.0
+ultralytics>=8.2.0
+
+# CLIP (OpenAI)
+ftfy>=6.2.0
+regex>=2024.5.15
+tqdm>=4.66.0
git+https://github.com/openai/CLIP.git
-# --- Utilities ---
-aiofiles>=23.2.1
-httpx>=0.26.0
-sentence_transformers
-openai-whisper
+# RAG / Document Ingestion
+pypdf>=4.3.0
+sentence-transformers>=3.0.0
+
+# Speech
+openai-whisper>=20231117
+
+# Utilities
+scipy>=1.13.0
diff --git a/simulator/requirements.txt b/simulator/requirements.txt
@@ -1,10 +0,0 @@
-fastapi
-uvicorn
-pydantic
-numpy
-torch
-Pillow
-python-dotenv
-azure-identity
-azure-digitaltwins-core
-openai
