1 of 11

Cloud Vision

AI-Powered Image Recognition App on AWS

A hands-on demo for TechDerby Cloud Enthusiasts.

2 of 11

The AI/Cloud Myth

Lots of people think cloud AI is harder than it actually is

Why I am here? to show that cloud and AI are more approachable than many people think by building a simple AI Serverless App on AWS Cloud

3 of 11

Architecture Overview

Serverless & Event-Driven

1. Upload: A user puts a photo into an S3 bucket.

2. Trigger: S3 sends an event to a Lambda function.

3. Analyze: Lambda calls Rekognition to "see" the image.

4. Log: The results are saved or printed to CloudWatch.

4 of 11

Object Detection

Identify thousands of objects (bikes, cars, people) and scenes (stadium, office) automatically.

Facial Analysis

Detect emotions, age ranges, and gender, or compare faces for identity verification.

Text Extraction

Extract printed or handwritten text from images (OCR) to automate data entry.

Meet AWS Rekognition

5 of 11

S3 Configuration: Create a bucket for image uploads and enable event notifications.

IAM Permissions: Grant your Lambda function 'rekognition:DetectLabels' and S3 'GetObject' permissions.

Lambda Function: Use Boto3, the AWS SDK for Python, to communicate with the services.

Building the App

IAM Roles Needed:

AmazonS3ReadOnlyAccess Allows Lambda to read images from S3
AmazonRekognitionReadOnlyAccess Allows Lambda to call Rekognition APIs
AWSLambdaBasicExecutionRole Allows Lambda to write logs to CloudWatch

The Trigger: Set up a "Put" event trigger so your lambda function runs automatically on every upload.

6 of 11

Results - Rekognition Labels

7 of 11

Amazon Rekognition App� Real-World Use

Cloud Architectures for Intelligent Image Indexing, Archiving, and Real-Time Security Response

8 of 11

E-COMMERCE CATALOGING

The Problem: Manual metadata entry for thousands of daily listings is slow and error-prone.

Automated Tagging: S3 uploads instantly trigger Rekognition to identify object classes, colors, and attributes.

Zero Data Entry: Metadata like "Red", "Sneaker", and "Lace-up" is auto-injected into search indexes.

Market Ready: Products become searchable in milliseconds, accelerating the time-to-sale for vendors.

9 of 11

INTELLIGENT MEDIA ARCHIVING

The Problem: Managing and searching through millions of unsorted visual assets for marketing or news.

Global Searchability: Rekognition processes bulk S3 dumps to create a rich metadata layer.

Natural Querying: Search for complex scenes like "Sunset AND Palm Tree" rather than browsing file folders.

Near Real-Time Indexing: New assets are processed as they land, ensuring the archive is always current.

10 of 11

INTRUSION DETECTION

The Problem: Conventional motion sensors trigger costly false alarms for animals, wind, or rain.

Intelligent Filtering: Lambda acts as a gatekeeper, only escalating events tagged as "Person" or "Vehicle".

Validated Alerts: Drastically reduce "Alert Fatigue" for security operators by ignoring environmental noise.

Immediate Action: High-confidence detections trigger instant SNS notifications to mobile emergency teams.