Capstone Project Stories

NirAmaya – AI-Powered Sustainable Healthcare NirAmaya uses advanced AI technologies including Generative AI (LLMs like Llama 3 via Groq), a RAG (Retrieval-Augmented Generation) pipeline, and speech AI (Web Speech API) to provide real-time, voice-based diagnosis in multiple Indian languages. It also leverages TensorFlow Lite for offline AI processing, ensuring low energy consumption and accessibility in remote areas. By reducing unnecessary travel and hospital load, NirAmaya promotes sustainable, eco-friendly healthcare delivery.

TerraQuant AI: Global Carbon Economy Platform - AI-powered carbon tracking, risk prediction, and optimization - Connects companies, investors, governments, and carbon projects - Features: - Carbon GPS: Live global carbon map - Climate Risk Index: Global standard for climate risk scoring - AI Fraud Detection: Prevents greenwashing and fake carbon credits - Predictive Compliance: Warns of regulatory changes - Carbon Asset Tokenization: Creates tradable digital assets - Empowers sustainable decisions, reduces risk, and builds trust in the carbon economy.

Developed a project for EV users that integrates hardware and software to deliver the following features.

• Real-time battery health monitoring with predictive degradation insights
• AI-powered detection of harmful charging and driving behaviors
• Optimized battery performance with accurate dynamic range prediction
• Extended battery lifespan and reduced total cost of ownership

Developed a project which helps the farmers and home gardeners to select the best fertilizer and spot to grow a healthy plant with the following features.

• Real-time camera and deep learning are used to detect plant diseases and plant health, and recommend suitable fertilizers.
• Soil and crop parameters (pH, N, P, K, moisture and health index) are analyzed using a Random Forest model to check fertilizer safety.
• Real-time camera analysis is used to monitor surrounding air and environmental effects on plant health and generate visual heatmap outputs.
• A web dashboard displays fertilizer recommendations, toxicity warnings and visual support to help improve crop productivity.

My name is Aditya Sunil Biranje, a Third-Year B.Tech student in Artificial Intelligence and Data Science at Dr. D. Y. Patil School of Science and Technology, Pune. As part of my academic journey and hands-on learning experience, my team and I are working on an innovative project titled “Grovia: Intelligence that Sustains Trust.” This project focuses on leveraging intelligent technologies to build reliable, data-driven solutions that promote trust, transparency, and efficiency in real-world systems. Through this project, I aim to explore the practical applications of AI and contribute towards building impactful and sustainable technological solutions.
Project Description:

• A blockchain-based system designed to track, verify, and trade carbon credits transparently
• Uses smart contracts (Solidity) to issue and manage carbon credit tokens (ERC-20 standard)
• Integrates Artificial Intelligence (Machine Learning) to verify environmental data before issuing credits
• Stores verification data securely using IPFS (decentralized storage)
• Uses Chainlink oracles to fetch real-world carbon pricing data
• Provides a user-friendly web interface (React/Next.js) for industries and organizations
• Ensures immutability, transparency, and fraud prevention in carbon credit systems.

Project Description:

• Artificial Intelligence and Machine Learning–based application for evaluating battery performance
• Analyzes battery performance based on parameters like voltage, current, temperature, and charge cycles
• Predicts key metrics such as battery capacity and State of Health (SOH)
• Determines whether a battery is suitable for reuse in applications like solar energy storage or should be recycled
• Built using Python, Google Colab, Scikit-learn, and Streamlit
• Developed an interactive web-based dashboard for user input and real-time predictions
• Helps identify degradation trends for better decision-making
• Recommends reuse options like solar storage or recycling
• Supports sustainable energy practices by reducing battery waste

An Agentic-AI sustainability intelligence system that transforms carbon accounting from reactive reporting to proactive optimization. Project Description:
Carbon Copilot is an enterprise-grade platform combining:

• Automated emission measurement
• Retrieval-Augmented Generation (RAG)
• LLM-powered reasoning
• Company-specific sustainability insights

Project Description:

• Helps electric vehicle users find nearby charging stations easily
• Real-time location tracking for accurate nearby station detection
• Uses GPS/location services to detect the user’s current location
• Integrates with map services like Google Maps
• Displays nearest EV stations on a map interface
• Shows important details:
  • Distance from current location
  • Charging type (fast/slow)
• Provides navigation/directions to the selected charging station
• User-friendly mobile/web interface with interactive map view
• Reduces range anxiety for EV users
• Promotes eco-friendly transportation and EV adoption

Project Description:

• Machine learning–based application for smart agriculture
• Recommends suitable crops based on soil nutrients and climatic conditions
• Predicts expected crop yield using the Random Forest algorithm
• Classifies crop suitability into Low, Medium, and High levels
• Assesses crop resistance to environmental conditions
• Built using Streamlit for an interactive and user-friendly interface
• Helps farmers make effective and data-driven agricultural decisions

An AI-powered tool that helps farmers in dry regions of North Karnataka plan climate-resilient livelihoods. By analyzing land, water availability, and skills, it recommends sustainable cash crops, inland fisheries, and alternative income options, ensuring financial stability and climate-smart farming practices.

PhishNet is an AI-driven real-time phishing website detection system that leverages machine learning (LightGBM) to identify malicious URLs with high accuracy and low latency. It includes a Chrome Extension that actively monitors browsing activity and alerts users instantly, enhancing digital security. By preventing phishing attacks, PhishNet contributes to sustainability by reducing financial fraud, protecting digital resources, and promoting a safer, more trustworthy online ecosystem.

• AI-based intelligent system for analyzing glacier melt and its impact on water security and flood risk
• Integrates real glacier data, climate variables, and machine learning models for accurate predictions
• Analyzes environmental factors such as temperature, precipitation, and solar radiation affecting glacier behavior
• Uses Explainable AI (XAI) to provide transparent and interpretable prediction results
• Converts glacier melt predictions into water runoff estimates and flood risk indicators
• Identifies regions vulnerable to water shortages and flood hazards at an early stage
• Built with an interactive web-based dashboard for visualization and analysis
• Supports decision-making for climate planning and resource management

Developed a Smart Waste Route Optimization System that helps maintain cleaner cities without causing disruptions to the public.

• The system monitors the fill levels of waste bins in real time and integrates with the city’s traffic data to identify the best routes and optimal timings for waste collection.
• This enables garbage collection teams to operate more efficiently and avoid congested areas.
• Additionally, the platform allows the public to report nearby waste issues, ensuring faster response and prioritized cleaning.
• Overall, the project promotes efficient waste management, improved urban hygiene, and a better quality of life for the community.

The system is designed as an end-to-end decision-support platform with the following features.

• The system predicts EV flexibility using advanced machine learning models and refines these predictions through physics constraints to ensure real-world feasibility.
• It dynamically scales power based on the battery’s State of Charge (SoC) and evaluates economic viability by considering battery degradation costs.
• Based on these factors, the system generates transparent and optimized V2G dispatch commands.
• An interactive dashboard visualizes the entire process, enabling users to clearly understand how decisions are made.
• By prioritizing not just accuracy but also transparency and trust, this solution supports smarter, safer, and more reliable integration of EVs into the energy grid.

Developed an intelligent EV solution that enhances real-world driving efficiency through seamless software integration.

• The system continuously monitors vehicle load in real time to provide accurate energy consumption insights, while dynamically correcting battery State of Charge (SoC) based on actual operating conditions.
• It also delivers precise range estimation by adapting to live usage patterns, helping users avoid uncertainty during travel.
• In addition, the platform offers a web-based charging station analysis tool and smart charging decision support, enabling EV users to identify optimal charging options and make informed decisions with confidence.

A smart, AI-powered system designed to assist farmers in making data-driven decisions by predicting:

• Crop yield prediction
• Crop price forecasting
• Alternate crop recommendations
• Soil nutrient deficiency analysis (NPK and pH)

• AI-based crop yield prediction using Random Forest
• Crop price forecasting using Linear Regression
• Real-time soil analysis using Arduino and moisture sensors
• Automated decision-making with actionable recommendations
• Instant report delivery via WhatsApp for easy farmer access

GeoSentinel AI is an AI-powered platform designed to monitor and prevent environmental challenges in Karnataka through real-time, data-driven intelligence.

The system integrates data from satellites, weather sources, government datasets, IoT sensors, and citizen reports into a unified platform, creating a live digital twin of Karnataka’s environment. It continuously tracks key indicators such as forest cover, lake health, groundwater levels, air quality, and urban heat. An AI prediction engine analyzes this data to identify risks early—for example, forecasting lake pollution, groundwater depletion, or ecological stress zones before they reach critical levels. A citizen reporting feature enables people to upload images or reports of environmental issues, which are validated using AI and fused with satellite data for improved accuracy.

GeoSentinel includes a prioritization layer that ranks environmental risks based on severity, impact, and data reliability, ensuring that decision-makers focus on the most critical issues. Beyond alerts, the system generates ready-to-use action plans, reports, and summaries, enabling faster and more effective responses. A built-in feedback mechanism tracks outcomes of interventions, allowing the system to continuously learn and improve. Additionally, it provides cost and policy insights, helping authorities understand the economic impact of immediate versus delayed action.

The project is designed to begin with a focused implementation and scale into a comprehensive environmental intelligence system.

Project Description:

• Machine learning–based application and browser extension for YouTube clickbait detection
• Analyzes video credibility based on metadata, titles, descriptions, and video transcripts
• Predicts clickbait probability using a Hybrid Ensemble model combining LightGBM (Statistical) and Llama 3 (LLM)
• Classifies video content into Misleading or Trustworthy levels
• Assesses and flags semantic manipulation or unfulfilled claims within videos
• Built using Flask for a robust backend API and JavaScript for a seamless, user-friendly Chrome extension interface
• Helps viewers avoid deceptive content and make informed, time-saving viewing decisions

The WattWise project was developed to address two converging problems: the lack of appliance-level energy awareness in standard household electricity bills, and the high carbon footprint associated with computationally heavy AI models. We designed and implemented a "Green AI" system that disaggregates monthly electricity bills into appliance-level consumption, detects wasteful usage, and estimates carbon footprints without relying on expensive IoT hardware or deep learning models that require massive computing power.

Developed a smart crop monitoring solution designed to be simple, reliable, and efficient in real-world conditions.

• The system operates through a direct mobile application, allowing farmers to easily access insights about their crops.
• Even in remote areas, the solution ensures uninterrupted functionality with offline operation, removing dependency on constant internet access.
• At its core, an efficient AI system analyzes crop conditions and provides meaningful insights to support better decision-making.
• Designed with simplicity in mind, the system offers an easy setup process, making it accessible to farmers without technical expertise.
• Its flexible hardware integration further ensures adaptability across different farming environments and needs.