Projects Category: AI, ML

Overview

Wasaq is an advanced GIS-based Endowment Management System designed to optimize the planning, development, and analysis of endowment lands. This intelligent web platform offers real-time data visualization, AI-driven insights, and secure investor interaction—all in one place.

Built with scalability, usability, and security in mind, Wasaq empowers endowment institutions and stakeholders to make data-driven decisions with confidence.

Key Features

• Interactive GIS Map: Visualize all endowment lands with real-time status updates using dynamic layers and filters.

• Advanced Analytics & Reporting: Analyze land use, performance trends, and development potential using big data and visual dashboards.

• AI-Powered Recommendations: Get smart suggestions for land development opportunities using machine learning models.

• Investor Support Chatbot: Engage potential investors with an AI-driven assistant built on NLP frameworks.

• Smart Filtering & Search: Seamlessly find properties and projects by location, status, category, or development stage.

• Economic Insights: Leverage big data analytics for economic forecasting and decision support.

Technology Stack

• Backend: Python, Flask

• Frontend: HTML5, CSS3, JavaScript, Bootstrap

• Database: PostgreSQL with PostGIS (for spatial data)

• GIS Tools: GeoPandas, Folium

• AI/ML: TensorFlow, scikit-learn, Transformers

• Chatbot: ChatterBot, NLTK, spaCy

• Data Visualization: Plotly, Matplotlib

Overview

The Medication Classification System is a powerful machine learning platform developed to automatically classify medications as either prescription or over-the-counter (OTC). Leveraging advanced Natural Language Processing (NLP) and transformer-based deep learning models, this tool assists medical professionals, researchers, and developers in understanding drug classifications quickly and accurately.

This project showcases a combination of AI, Django web development, and automated PDF processing, delivering a complete end-to-end solution for analyzing medical texts.

How It Works

1. Input

Users can either:

• Upload PDF documents containing medication leaflets or packaging information.

• Enter the medication name directly into the platform.

2. Text Extraction (For PDF)

Our system uses PyPDF2 to:

• Extract raw content from the uploaded file

• Identify critical drug details such as:

  • Medication Name

  • Active Ingredients

  • Dosage Form & Strength

3. Preprocessing & Analysis

The extracted or entered data undergoes:

• Text normalization (lowercasing, cleaning)

• Stopword removal

• Lemmatization

• Tokenization with RoBERTa tokenizer

4. AI-Powered Classification

Once processed, the information is passed into fine-tuned NLP models:

• RoBERTa (primary model)

• BERT and DistilBERT (alternative and comparative models)

The system returns:

• Classification: Prescription or OTC

• A confidence score for each prediction

Technical Stack

• Language: Python
• Backend Framework: Django
• AI Models: RoBERTa, BERT, DistilBERT
• NLP Libraries: HuggingFace Transformers, NLTK
• PDF Processing: PyPDF2
• Frontend: Django Templates

Use Cases

• Healthcare Systems – Automate drug classification workflows

• Medical Research – Analyze large datasets of drug literature

• Pharma HR & Legal – Ensure regulatory compliance on medication labeling

• HealthTech Apps – Integrate medication classification in mobile/web platforms

Grass is a smart mobile application designed to transform modern agriculture by offering AI-powered recommendations to farmers and agricultural professionals. With the global population rising and natural resources becoming increasingly scarce, there’s a critical need for intelligent systems that improve crop productivity, ensure sustainability, and reduce waste. Grass addresses these challenges by providing real-time crop recommendations, plant disease detection, and smart data-driven insights, all from a single, user-friendly mobile platform.

Why Grass?

Traditional farming methods often lack efficiency due to limited data, climate variability, and unpredictable soil conditions. Grass leverages Machine Learning (ML) and Computer Vision to make farming smarter, more accurate, and future-ready. This solution empowers users—especially farmers—with personalized crop suggestions, disease prevention advice, and ongoing support powered by AI.

Key Features and Functional Requirements

1. User Management

• Account Registration & Login: Secure sign-up and sign-in via email and password.

• Profile Management: Users can view and edit their agricultural data history.

• Admin Dashboard: Allows role management and data control to maintain security and system integrity.

2. Crop Recommendation System

• Input Soil Data: Enter soil pH, moisture, and nutrient levels.

• AI-Based Analysis: Uses a Decision Tree algorithm to analyze soil characteristics and environmental factors.

• Smart Suggestions: Recommends the most suitable crops along with optimal planting conditions and timing.

• Dynamic Updates: Adjusts recommendations in real time based on new data.

3. Plant Disease Prediction

• Image Upload: Users can upload images of diseased leaves or plants.

• CNN-Based Diagnosis: Utilizes Convolutional Neural Networks to identify plant diseases.

• Treatment Guidance: Provides detailed information on symptoms, causes, and actionable treatments.

• Admin Tools: Admins can update the disease library to keep predictions relevant and up-to-date.

4. Data & Analytics

• Real-Time Sync: Uses Firebase for fast data storage and retrieval.

• Admin Dataset Management: Enables dynamic updates of crop and disease datasets.

• Usage Reports: Helps track trends, system performance, and improve future recommendations.

5. User Interface & Usability

• Flutter-Based Cross-Platform UI: Ensures responsive design on Android and iOS.

• Guided Forms: Streamlines the data input process with structured forms and prompts.

• Visual Feedback: Presents results in visually intuitive formats including charts, images, and text.

• Error Handling: Real-time alerts for data input errors or upload issues.

Behind the Tech: Algorithms & Architecture

1. Machine Learning Techniques

• Decision Trees: For intelligent crop recommendation.

• CNN (Convolutional Neural Networks): For high-accuracy disease prediction based on plant image inputs.

• Data Normalization & Cleaning: Ensures high-quality inputs for model training and predictions.

• Model Training: Uses updated agricultural datasets for better prediction accuracy over time.

2. Technologies & Frameworks

• Flutter: Cross-platform UI development

• Firebase: Authentication, real-time database, and cloud storage

• Python: Backend processing and ML algorithm integration

• TensorFlow/Keras: For model development and deployment

• OpenCV & Torchvision: For image preprocessing and plant disease detection

• Scikit-Learn: For classic ML tasks like classification and decision tree modeling

App Architecture & Integration

Grass is built using a modular multi-tier architecture that ensures smooth communication between the frontend, backend, and AI components.

Mobile App (Client Layer)

• Built using Flutter for responsive design

• Integrates Firebase Authentication

• Uses Room SQLite for local caching and offline mode

• Supports secure image uploads and soil data input

Backend Services (Processing Layer)

• Python-based backend handles logic for ML predictions

• Communicates with Firebase for real-time syncing

• Ensures scalable deployment and easy maintenance

ML/AI Engine (Model Layer)

• Pre-trained CNN models using TensorFlow/Keras for plant disease recognition

• Decision Tree models for crop recommendation

• Real-time inference capabilities with low latency

In an increasingly connected digital world, cyber threats continue to evolve at an alarming rate. With phishing attacks and malicious links becoming more sophisticated, there is an urgent need for intelligent, real-time protection systems. That’s where this AI and Machine Learning-powered Malicious URL Detection System comes into play—a cutting-edge solution designed to help users stay protected from online threats before they strike.

This project is a robust mobile and backend system developed using modern technologies like Flask (Python), TensorFlow, Firebase, and Flutter, with a core focus on real-time threat detection, speed, security, and usability.

Project Goals

The primary objective of this project is to safeguard users from phishing, malware, and other cyber threats by analyzing URLs in real-time using artificial intelligence and machine learning models. It empowers users with instant threat detection, historical scan tracking, and cybersecurity education.

Key Features & Functionalities

URL Safety Analysis

• Real-Time Detection: Analyze links using a trained LSTM model to detect phishing, malware, and suspicious patterns.

• AI-Based Classification: Uses a deep learning model trained on malicious and safe URL datasets for high accuracy and fast inference.

Scan Results & Insights

• Detailed Scan Feedback: Displays if the URL is safe or malicious, with context on detected risks.

• Scan History: Users can view their previously scanned links with timestamps, helping them stay aware of recurring threats.

Instant Notification Alerts

• Users are immediately notified if a link is classified as dangerous, preventing potential harm before any interaction.

Mobile Application Design

Built with a modern Android architecture, the mobile app is user-centric and secure:

Security-Focused Login

• OAuth 2.0 & JWT Authentication

• Firebase for credential management

• Encrypted session storage

Intuitive User Interface

• URL submission with regex validation

• History view powered by SQLite (Room)

• URL caching with LRU policy for speed

• Dark/light themes and accessible UI

Machine Learning & Model Deployment

LSTM-Based Detection Engine

• Bidirectional LSTM Model with ONNX runtime for performance

• Achieves 98.4% F1-score using datasets from PhishTank and VirusTotal

• Handles requests in 320ms on CPU, 85ms on GPU

Preprocessing & Tokenization

• Custom tokenizer supports Unicode-safe encoding

• Sequences converted for LSTM input with max length of 200

Model Serving

• Deployed using TensorFlow Serving over gRPC

• Supports async requests via Flask API

Backend Architecture

A multi-tier microservices architecture ensures scalability and high performance:

Flask API Gateway

• Blueprint routing and rate limiting (100 req/IP/hr)

• Input sanitization with Bleach to prevent XSS/SQLi

• Secure API request handling using HMAC verification

API Endpoints

• POST /api/v1/scan: Accepts a URL and returns classification

• GET /api/v1/history: Fetches user’s scan history

Security Features

• HTTPS encryption for all client-server communication

• JWT-based session management

Technical Stack

• Frontend (Mobile): Flutter

• Backend API: Flask (Python)

• Authentication: Firebase, OAuth 2.0, JWT

• Machine Learning: TensorFlow, LSTM, ONNX

• Storage: Firebase, SQLite

• Communication: REST + gRPC