Potato Leaf Disease Detection

End-to-End AI Solution for Agricultural Sustainability

Project Overview

"From Farm Crisis to AI-Powered Prevention"

Potato leaf diseases cost farmers millions globally through crop losses. This project transforms agricultural disease management by deploying a full-stack deep learning solution that predicts Early and Late Blight with clinical precision, accessible through both web and mobile interfaces.

🎯 Problem Impact

• Economic Loss: Early & Late Blight cause devastating crop failures worldwide

• Detection Gap: Manual disease identification is slow, subjective, and often too late

• Accessibility: Farmers need instant, accurate diagnosis tools in the field

💡 Solution Architecture

"Intelligence at Every Layer"

AI Core: CNN-based image classification with 95%+ accuracy Backend: FastAPI for lightning-fast asynchronous processing
Frontend: React with Material-UI for intuitive user experience Mobile: React Native for real-time field deployment Cloud: GCP Functions for scalable, global accessibility

🔬 Technical Deep Dive

Deep Learning Model

• Architecture: Convolutional Neural Network (CNN) with ReLU activation

• Dataset: Kaggle PlantVillage dataset with image preprocessing and augmentation

• Training: Model compilation, training, and evaluation for disease classification

• Performance: Achieved strong accuracy in distinguishing between healthy and diseased potato leaves

• Export: Saved trained model (.h5 format) for deployment

API Development

• FastAPI: Built backend API to serve the trained model

• Model Loading: Integrated saved .h5 model for making predictions

• Image Processing: Created endpoint to receive and process uploaded images

Frontend Development

• React + Material-UI: Built user interface for image upload and prediction display

• Image Processing: Integrated file upload functionality with prediction results

• User Interface: Created simple web interface for model interaction

Cloud Deployment

• GCP Setup: Custom project with dedicated storage bucket

• Model Hosting: .h5 model deployment via Cloud Functions

• API Endpoint: https://us-central1-leaf-disease-classification-1.cloudfunctions.net/predict

• Testing: Postman validation ensuring production readiness

Mobile Solution

• React Native: Cross-platform mobile application

• Android Emulator: Full testing environment via Android Studio AVD

• Real-time Processing: Live camera integration with instant disease prediction

📊 Key Achievements

Technical Implementation:

• CNN-based disease classification model

• Web interface for image upload and prediction

• Cloud deployment via Google Cloud Platform

• Mobile app testing using Android emulator

Learning Achievements:

• Built and trained CNN model from scratch

• Deployed deep learning model to cloud platform

• Created functional web interface for model interaction

• Integrated multiple technologies for complete solution

🛠️ Technology Stack

• AI/ML: Python, TensorFlow/Keras, CNN, Transfer Learning 

• Backend: FastAPI, Python, Async Programming 

• Frontend: React, Material-UI, Axios, Responsive Design 

• Mobile: React Native, Android Studio, Emulator Testing 

• Cloud: Google Cloud Platform, Cloud Functions, Storage Buckets 

• Testing: Postman, Production Validation 

• Development: Jupyter Notebook, PyCharm, VS Code

🚀 Deployment Pipeline

1. Model Training: CNN architecture with optimized hyperparameters

2. API Development: FastAPI backend with async capabilities

3. Web Interface: React frontend with intuitive UX

4. Cloud Deployment: GCP Functions for global scalability

5. Mobile App: React Native for field accessibility

6. Testing: Comprehensive validation across all platforms

📈 Business Impact

For Farmers:

• Instant disease diagnosis in the field

• Prevention-focused crop management

• Reduced economic losses from disease outbreaks

For Agriculture:

• Scalable AI solution for crop health monitoring

• Data-driven disease prevention strategies

• Sustainable farming through early intervention

🔗 Project Resources

Live Demo: YouTube Demonstration 

Source Code: GitHub Repository 

API Endpoint: Production-ready GCP deployment

📋 Learning Journey

"Building My First Deep Learning Application"

This project represents my foundational step into deep learning—focusing on understanding CNN architecture, model training, and deployment. Through guided learning and tutorials, I developed:

• CNN model development and training

• Basic web interface creation

• Cloud deployment fundamentals

• Integration of machine learning with web technologies

Approach: "Learn | Build | Deploy" - focusing on core deep learning concepts while exploring supporting technologies.

🎯 Portfolio Highlight

This project demonstrates:

• Deep Learning Foundation: CNN implementation for image classification

• Practical Application: Addressing real agricultural challenges

• End-to-End Development: From model training to deployment

• Technology Integration: Combining AI with web and cloud technologies

Result: A foundational deep learning project that showcases technical learning ability and practical problem-solving approach.

AI-Powered PCB Defect Detection System

Transforming Quality Control Through Machine Vision

🎯 Project Overview

In the microscopic world of electronics manufacturing, a single defect smaller than a grain of sand can render an entire circuit board useless. This project tackles one of the industry's most challenging problems: detecting sub-millimeter defects in Printed Circuit Boards (PCBs) that human inspectors often miss until the third examination pass.

Client: Flextronics Technology India Pvt. Ltd., Bengaluru
Executing Partner: TANSAM
Team Lead: AI Developer

🚀 The Challenge: When Human Eyes Aren't Enough

Traditional PCB inspection in repair stations relies on technicians with magnifying glasses—a process that's:

• Time-intensive: Multiple inspection passes required

• Error-prone: Microscopic defects escape detection

• Costly: Defective boards reach customers, causing field failures

• Inconsistent: Quality varies with inspector fatigue and experience

The stakes? A single missed defect can cost thousands in warranty claims and damage brand reputation.

💡 The Solution: Machine Vision That Never Blinks

We developed an AI-powered quality inspection system that combines:

Core Technologies

• Computer Vision: Advanced image processing algorithms

• Deep Learning: CNN-based defect classification

• Anomaly Detection: Siamese networks for golden-board comparison

• Transfer Learning: Leveraging pre-trained models for enhanced accuracy

Key Innovation: Tile-Based Golden Reference

Instead of traditional broad segmentation, our system:

1. Divides PCBs into micro-tiles

2. Compares each tile against a "golden board" reference

3. Flags deviations at sub-millimeter precision

4. Provides exact defect coordinates and classifications

🛠️ Technical Architecture

Hardware Integration

• 360° Rotating Camera System: Single fixed camera captures comprehensive board views

• Automated Inspection Station: Integrated workstation for seamless workflow

• Cloud-Ready Infrastructure: SharePoint integration for enterprise deployment

AI Model Pipeline

Image Capture → Preprocessing → Tile Segmentation → 

CNN Classification → Anomaly Detection → 

Golden Board Comparison → Defect Localization → 

Quality Report Generation

Defect Detection Capabilities

• PCB-Level Defects: Water damage, corrosion, pad damage, missing components

• RRU-Unit Defects: Missing screws, bent pins, connector damage, air-leak caps

• Microscopic Defects: Sub-millimeter faults invisible to naked eye

📊 Project Development Journey

Phase 1: Initial Model Development

• Created baseline CNN model with limited accuracy (<50%)

• Implemented data augmentation techniques to expand dataset

• Tested multiple architectures: VGG16, YOLO, Auto-encoders, Isolation Forest, DBSCAN

Phase 2: Problem Analysis & Client Engagement

• Site Visit to Flextronics Bangalore (January 9, 2024)

• Observed real-world inspection challenges and workflow

• Identified 30+ distinct PCB patterns requiring detection

• Gathered requirements for microscopic defect detection

Phase 3: Technical Pivot

• Shifted from broad segmentation to tile-based comparison methodology

• Implemented Siamese networks for anomaly detection

• Developed golden-board reference comparison system

• Experimented with various deep learning approaches

Phase 4: MVP Development

• Created functional prototype using Streamlit framework

• Demonstrated core defect detection capabilities

• Integrated basic image processing and classification features

• Delivered minimum viable product for client evaluation

🔧 Technical Challenges & Learnings

Key Challenges Encountered

• Limited Dataset: Initial lack of diverse defective PCB samples

• Model Accuracy: Difficulty achieving production-ready accuracy levels

• Microscopic Detection: Complex requirements for sub-millimeter defect identification

• Real-world Constraints: Adapting academic approaches to industrial requirements

Technical Approaches Tested

• Convolutional Neural Networks (CNN): Base architecture for image classification

• Data Augmentation: Synthetic data generation to expand training dataset

• Siamese Networks: Twin network architecture for similarity comparison

• Golden Board Comparison: Reference-based anomaly detection methodology

• Multiple Algorithms: VGG16, YOLO, Auto-encoders, Isolation Forest, DBSCAN

MVP Implementation

• Developed functional prototype using Streamlit framework

• Integrated basic image upload and processing capabilities

• Implemented defect classification for common PCB issues

• Created user-friendly interface for demonstration purposes

💼 Professional Skills Demonstrated

Technical Competencies

• Machine Learning: Model development, training, and evaluation

• Computer Vision: Image processing and feature extraction

• Deep Learning: CNN architectures and neural network optimization

• Data Science: Dataset creation, augmentation, and analysis

• Software Development: Streamlit application development

• Problem Solving: Iterative approach to complex technical challenges

Project Management

• Client Engagement: On-site visits and requirement gathering

• Stakeholder Communication: Technical discussions with engineering teams

• Research & Development: Systematic evaluation of multiple approaches

• Documentation: Comprehensive project tracking and reporting

📈 Key Learnings & Insights

1. Domain Knowledge: Understanding manufacturing processes is essential for effective AI implementation

2. Data Quality: High-quality, diverse datasets are critical for model performance

3. Iterative Development: Multiple approaches and continuous refinement are necessary

4. Client Collaboration: Regular feedback and on-site visits provide valuable insights

5. Realistic Expectations: Complex problems require significant time and resources to solve effectively

🔗 Technologies Used

Machine Learning: TensorFlow, PyTorch, Scikit-learn
Computer Vision: OpenCV, PIL
Deep Learning: CNN, Siamese Networks, Transfer Learning
Development: Python, Streamlit
Data Processing: NumPy, Pandas
Visualization: Matplotlib, Seaborn

This project provided valuable experience in applying machine learning to real-world manufacturing challenges, demonstrating the complexity of developing production-ready AI systems for industrial applications.

HR Data Analytics Dashboard

Transforming Raw Employee Data into Strategic Insights

🎯 Project Overview

Human Resources departments often struggle with fragmented employee data scattered across multiple Excel files, making it difficult to spot trends, identify concerns, or make informed decisions quickly. This project addressed that challenge by creating a comprehensive HR analytics dashboard that transforms raw employee data into actionable insights.

Data Source: Excel-based HR files
Platform: Microsoft Power BI
Project Type: Solo development
Target Users: HR managers, executives, and department heads

🔍 The Business Challenge

HR teams needed a centralized view to answer critical questions:

• What is our current workforce composition?

• How are salaries distributed across different roles and qualifications?

• Are there concerning patterns in leave balances?

• How has our team grown over time?

• What demographic trends should we be aware of?

Without proper analytics, these questions required hours of manual Excel work, often leading to outdated or incomplete insights.

🛠️ Technical Implementation

Data Foundation & Power Query

The project began with data preparation challenges in Power Query:

• Data Type Corrections: Converted text fields to proper date and numeric formats

• Educational Qualification Transformation: Created numeric coding system for qualifications to enable scatter plot analysis

• Data Validation: Cleaned inconsistent entries and standardized naming conventions

• Age Grouping: Implemented binning logic to categorize employees into meaningful age ranges

Data Modeling Architecture

• Relationship Design: Established connections between employee data and qualification dimension tables

• Qualification Dimension: Created separate table for educational qualifications with numeric IDs

• Date Intelligence: Implemented proper date handling for time-based analysis

• Data Integrity: Ensured referential integrity across related tables

Advanced DAX Measures

Developed custom measures beyond basic aggregations:

Head Count = COUNTROWS(Employees)

Average Salary = AVERAGE(Employees[Salary])

Min Salary = MIN(Employees[Salary])

Max Salary = MAX(Employees[Salary])

Cumulative Head Count = 

    CALCULATE(

        [Head Count],

        FILTER(

            ALLSELECTED(Employees[Date of Join]),

            Employees[Date of Join] <= MAX(Employees[Date of Join])

        )

    )

Avg Leave Balance = AVERAGE(Employees[Leave Balance])

High Leave Count = 

    CALCULATE(

        [Head Count],

        Employees[Leave Balance] > 20

    )

📊 Dashboard Components

Core Metrics Cards

• Headcount (161): Total employee count with dynamic filtering

• Average Leave Balance (16 days): Monitoring leave utilization

• High Leave Alert (29 employees): Flagging potential burnout cases

• Average Salary ($54K): Compensation overview

Analytical Visualizations

1. Job Title Distribution: Horizontal bar chart showing workforce composition

2. Gender Demographics: Pie chart revealing 55% female workforce

3. Growth Trajectory: Line chart displaying cumulative headcount over time

4. Age Distribution: Histogram showing "young and rippen" workforce pattern

5. Salary vs. Qualification Analysis: Scatter plot exploring compensation correlations

Interactive Features

• Job Title Slicer: Filter entire dashboard by specific roles

• Dynamic Filtering: Cross-filtering between all visualizations

• Drill-down Capabilities: Navigate from summary to detailed views

• Contextual Tooltips: Additional information on hover

💡 Key Insights Discovered

Hidden Patterns Revealed

• Gender-Age Distribution: Uneven distribution across age groups requiring attention

• Salary Ranges by Role: Significant variations within job categories

• Leave Balance Concerns: 18% of staff (29 employees) with excessive leave accumulation

• Growth Trends: Steady workforce expansion with acceleration in recent years

Actionable Intelligence

• Compensation Analysis: Identified roles with compressed salary ranges

• Leave Management: Highlighted employees needing leave counseling

• Workforce Planning: Clear growth trajectory supporting future planning

• Diversity Insights: Strong female representation across the organization

🔧 Technical Challenges & Solutions

Data Transformation Hurdles

• Challenge: Educational qualifications as text couldn't be used in scatter plots

• Solution: Created numeric coding system through Power Query transformations

• Challenge: Inconsistent date formats across Excel files

• Solution: Standardized date parsing with error handling

Advanced Analytics Implementation

• Age Binning: Grouped continuous age data into meaningful categories

• Running Totals: Implemented cumulative calculations for growth analysis

• Top-N Filtering: Dynamic filtering for high-priority metrics

• Relationship Mapping: Connected disparate data sources through common keys

📈 Business Impact

Decision-Making Enhancement

• Instant Access: Stakeholders can now access HR metrics in seconds rather than hours

• Trend Recognition: Clear visualization of workforce patterns and growth trajectories

• Proactive Management: Early identification of issues like excessive leave accumulation

• Strategic Planning: Data-driven insights supporting long-term HR strategies

Operational Efficiency

• Reduced Manual Work: Eliminated repetitive Excel analysis tasks

• Consistent Reporting: Standardized metrics across all HR communications

• Real-time Insights: Up-to-date information supporting timely decisions

• Cross-functional Access: Multiple stakeholders using the same reliable data source

🎨 Design & User Experience

Visual Design Principles

• Clean Layout: Organized information hierarchy with clear visual separation

• Consistent Branding: Professional color scheme and formatting

• Intuitive Navigation: Logical flow between different analytical views

User-Centric Features

• Interactive Slicers: Easy filtering without technical knowledge required

• Dynamic Titles: Context-aware headings that update with filters

• Clear Labeling: Descriptive titles and axis labels for immediate understanding

• Tooltip Enhancement: Additional context without cluttering main visuals

🔗 Technical Skills Demonstrated

Power BI Expertise:

• Power Query data transformation and cleansing

• DAX measure creation and calculated columns

• Data modeling and relationship management

• Advanced visualization techniques

• Interactive dashboard design

Analytical Capabilities:

• Statistical analysis and trend identification

• Data grouping and binning strategies

• Time-series analysis with running totals

• Correlation analysis through scatter plots

• Demographic and compensation analytics

Data Management:

• Excel data integration and standardization

• Data quality assessment and improvement

• Dimension table creation and management

• Cross-table relationship establishment

📚 Lessons Learned

1. Data Preparation is Critical: Quality insights require quality data preparation

2. User-Centric Design: Dashboard success depends on end-user adoption and usability

3. Iterative Development: Regular stakeholder feedback improves final outcomes

4. Context Matters: Raw numbers need contextual analysis for meaningful insights

5. Visual Hierarchy: Proper layout guides users to the most important information

This HR analytics dashboard demonstrates the power of transforming scattered Excel data into a centralized, interactive platform that empowers data-driven decision making across the organization.