AI-Assisted Web Application

Applied AI portfolio project

CropGuard AI

A crop disease decision-support application combining image classification, prediction history, user accounts, feedback workflows, an AI assistant, and administrator tools in a responsive Flask interface.

My role

Full-stack & ML integration developer

Core stack

Python · Flask · TensorFlow

View source Live demo

portfolio / projects / plant-disease-detection

The problem

Crop symptoms can be difficult to identify consistently from visual inspection alone, while a raw machine-learning notebook does not provide the workflow, history, guidance, or usability needed by non-technical users.

The solution

I connected the model workflow to a responsive Flask application with authentication, upload/camera support, prediction history, decision-support messaging, feedback collection, administration tools, and database-backed user workflows.

Project overview

From practical need to complete product.

CropGuard AI connects a trained image-classification workflow to a practical web application. Users can upload or capture crop images, receive model predictions and guidance, review detection history, and interact with supporting tools while administrators manage feedback and application data.

PythonFlaskTensorFlowKerasPostgreSQLJavaScript

Responsive product view

One interface, every screen.

The same product is presented across desktop, tablet, and phone frames to show how the interface adapts, not just how it looks in one screenshot.

desktop

tablet

phone

System architecture

A clear delivery path from the user interface through application logic to data and external services.

01Responsive Flask/Jinja user interface

02Python application and prediction pipeline

03TensorFlow/Keras image-classification model

04Relational database for users, history, and feedback

05Email/reset and deployment-ready service integrations

My responsibilities

Model-to-web integration and image-processing workflow
User authentication, history, feedback, and admin functionality
Responsive light/dark interface and mobile camera workflow
Database and deployment configuration

Key outcomes

Moved an image-classification model into a usable end-to-end application
Added history, feedback, admin, and guidance around model predictions
Designed responsive upload and camera workflows for desktop and mobile

Key features

Image upload and mobile camera prediction workflow
Prediction history and user accounts
Admin feedback management with CSV export
Responsive AI guidance interface
Decision-support messaging for uncertain predictions

Engineering challenges

Improving generalization beyond the original training images
Connecting model preprocessing consistently to web uploads
Keeping AI guidance usable on small screens
Balancing prediction confidence with responsible user messaging

Future improvements

Retrain with a larger and more diverse field-image dataset
Add calibrated confidence and out-of-distribution detection
Expand crop and disease coverage
Add expert-reviewed treatment guidance

Lessons learned

What this project improved in my engineering process.

01Model accuracy and product usability must be evaluated separately
02AI predictions need clear uncertainty and decision-support language
03Real-world testing data is essential before presenting a classifier as reliable

Next case study

E-Millenial Store

A responsive e-commerce experience with product browsing, cart management, customer validation, purchase summaries, and Paystack checkout integration.

Building the experience

Loading projects, engineering decisions, and product details.

Back to projects