Cardiolytics

Project Overview

Cardiolytics: Cardiovascular Analytics for Homecare

Cardiolytics is a web-based cardiovascular analytics platform that integrates IoT, ensemble learning, and a document-grounded chatbot to support cardiovascular homecare services. The system consists of three core features:

Core Features

BP Monitor – Real-time Blood Pressure Monitoring

This feature uses an ESP32 microcontroller to simulate real-time blood pressure data transmission via MQTT (EMQX Broker) and Apache Kafka, enabling:

• Device connectivity via the Connect to Device button
• Measurement trigger through Start Measurement
• Live dashboard displaying systolic, diastolic, and heart rate data
• Automatic detection of connection status (connected/disconnected)
• Secure data persistence into a backend MySQL database via Flask API

CVD Predict – Cardiovascular Disease Risk Prediction

Users can assess their risk of cardiovascular disease by submitting clinical and lifestyle data. The modeling pipeline includes:

• Two-phase Exploratory Data Analysis (EDA)
• Preprocessing (column renaming, IQR-based outlier removal, feature encoding)
• Feature engineering:
  • Body Mass Index (BMI)
  • Pulse Pressure (PP)
  • Mean Arterial Pressure (MAP)
• Model training using:
  • Gradient Boosting (LightGBM, CatBoost, HistGB)
  • Tabular Deep Learning (TabNet, TabTransformer, FT-Transformer)
  • Stacking & Hybrid Voting Ensembles
• Best model (Voting Hybrid) achieves:
  • Accuracy: 74.08%
  • AUC: 0.8018
• Final prediction reports are downloadable in PDF format

Cardiobot – Document-Based Educational Chatbot (RAG)

This Retrieval-Augmented Generation (RAG) chatbot answers cardiovascular-related questions based on uploaded verified documents:

• Uses Gemini 2.5 Flash as the main LLM
• Embeds medical documents via multilingual-e5-large
• Stores chunks in Pinecone vector store
• Supports conversational memory via LangChain
• Validates answers using:
  • Cosine similarity
  • Word overlap detection
• Caches repeated questions to optimize performance
• Links answers to the exact source document section

Web Interface

Cardiolytics is accessible via a web interface, structured for two user roles:

User Side (Patients)

• View health summary and connected device status
• Submit predictions through the CVD Predict form
• View prediction history and download medical PDF reports
• Chat with Cardiobot for document-grounded health information

Admin Side

• Approve new patient registrations
• Monitor all users’ BP and CVD prediction data
• Manage model uploads (.pkl), IoT devices, and chatbot documents
• Preview document content

Key Achievements

• BP Monitor: Real-time monitoring pipeline with MQTT → Kafka → Flask integration, powered by simulated ESP32 data
• CVD Predict: Achieved 74.08% accuracy and 0.8018 AUC using hybrid ML/DL ensemble
• Cardiobot: Gemini-powered RAG chatbot with document-based Q&A and source-linked responses
• Web App: Role-based interface for patient and admin
• Deployment: Fully tested locally with end-to-end data flow validation