System Overview

Our AI-driven solution is a low-cost, plug-and-play water usage tracker and intelligent advisory system that enables households to:

• Monitor real-time water consumption via a sensor attached to water taps.
• Detect leaks and anomalies to reduce wastage.
• Analyze water quality based on pH and TDS values.
• Receive personalized recommendations on sustainable water practices.
• Access globally recognized water guidelines from the UN and WHO.
• Receive multilingual AI-generated insights and alerts for accessibility.

It leverages:

• IBM Cloudant for schema-less NoSQL storage
• IBM Event Stream for real-time streaming of water consumption data into Cloudant
• ChromaDB for high-dimensional vector embeddings
• IBM WatsonX (LLMs) for natural language question-answering
• Streamlit for interactive UI

---

Harsware setup used for prototyping

Visual representation data ingestion

High-Level Architecture Diagram of Agentic AI

---

Component-Wise Breakdown

Sensor Layer

• rpi_connector.py: Collects raw water quality metrics from Raspberry Pi interfacing with the sensor.
• Directly reads from hardware GPIO/UART.

---

Data Ingestion Layer

• IBM Event Stream is used as the real-time streaming backbone, via Kafka.
  • kafka_producer.py: Publishes real-time data into Kafka topics.
  • kafka_consumer.py: Subscribes to the topic and processes the streaming data.

---

Data Persistence Layer

NoSQL Storage

• IBM Cloudant:
  • A scalable, distributed, document-oriented database service.
  • Used to store ingested water data as JSON documents.
  • Scripts:
    • load_cloudant_in_json.py: Upload JSON to Cloudant.
    • read_cloudant.py: Retrieve and read from Cloudant.
    • cloudant_search.py: Query using Cloudant’s built-in Mango query language.

Local Caching

• SQLite DB:
  • json_to_sqlite.py: Converts JSON documents from Cloudant into SQLite for faster local querying and backup for this MVP.

---

Vector Database Layer

• ChromaDB (Open-source vector store):
  • Converts unstructured water/environmental data into dense vector embeddings.
  • load_vector_un.py: Vectorizes and stores using SentenceTransformers/Text2Vec.
  • read_vector.py: Validates vector retrieval and relevance.

---

User Interface Layer

• Streamlit Web App:
  • system_chat.py is the entry point.
  • Enables users to:
    • Ask natural language questions (e.g., "Is the water quality below threshold?")
    • View intelligent answers with explanations.
    • Visualize results (can be extended with charts or tables).

---

Technologies Used

Tech	Description
IBM Cloudant	Managed NoSQL database for JSON documents.
Apache Kafka	Distributed messaging for real-time streaming.
SQLite	Lightweight SQL database for local analysis.
ChromaDB	Vector DB for storing and retrieving document embeddings.
WatsonX LLMs	IBM's foundation models for natural language understanding.
Streamlit	Lightweight UI for interacting with agents.
LangChain	Agent chaining, memory, and prompt templates.
Python	Core programming language for scripting and orchestration.

Folder Structure

Folder/File	Description
kafka_producer.py	Sends sensor data to IBM Event Stream.
kafka_consumer.py	Receives Kafka messages and insert that in Cloudant DB.
load_cloudant_in_json.py	Pushes JSON data into Cloudant.
read_cloudant.py	Reads from Cloudant DB.
cloudant_search.py	Runs Cloudant queries.
json_to_sqlite.py	Converts Cloudant JSON to SQLite.
swarm_agents.py	Multiple agents collaborating to answer queries.
load_vector_un.py	Load document data into Chroma vector DB.
read_vector.py	Retrieve and validate vector embeddings.
system_prompt.py	Prompt templates for AI agents.
system_chat.py	Chatbot interface using Streamlit.

---

Setup Instructions

1. Clone the Repo

git clone <repo-url>
cd ibm_hackathon

2. Setup Environment Variables

Create a .env file and add:

CLOUDANT_APIKEY=
CLOUDANT_URL=
DATABASE_NAME=

WATSONX_APIKEY=
WATSONX_PROJECTKEY=

TWILIO_SID=
TWILIO_AUTH_TOKEN=
TWILIO_WHATSAPP_NUMBER=
MY_WHATSAPP_NUMBER=

LANGCHAIN_API_KEY=

KAFKA_BOOTSTRAP_SERVER=
KAFKA_SSL_PASSWORD=

TAVILY_API_KEY=

3. Install Dependencies

pip install -r requirements.txt

---

Run the Project

Step 1: Load Data to Cloudant

python load_cloudant_in_json.py

Step 2: Convert to SQLite

python json_to_sqlite.py

Step 3: Populate Vector DB

python load_vector_un.py

Step 4: Validate Vector Storage (Optional)

python read_vector.py

Step 5: Run the Agent Chat UI

streamlit run system_chat.py

---