AWS Serverless Image Analysis Pipeline

A fully automated serverless pipeline that analyzes uploaded images using AWS Rekognition and stores the results in DynamoDB for easy retrieval.

---

The Challenge

Ever tried organizing thousands of photos on your phone? You know the struggle.
You've got pictures of your dog, your coffee, random screenshots, and vacation photos all mixed together. Finding anything takes forever.

Now imagine you're a business that gets hundreds of images uploaded daily - product photos, user content, documents. Someone has to manually look at each one and figure out what's inside.

That's exactly what I wanted to solve:

• Stop the manual work - No more humans staring at images all day
• Get instant answers - What's actually in this image?
• Make it searchable - Find all images with "cars" or "people" in seconds
• Never worry about scale - Whether it's 10 images or 10,000, it should just work

---

My Steps

1. Setting Up the Storage Layer
I created an S3 bucket to store incoming images and a DynamoDB table to hold analysis results.

• S3 handles the image storage and triggers
• DynamoDB stores structured analysis data with ImageName as the primary key

---

2. Building the Lambda Function
The core logic lives in a Python Lambda function that:

• Listens for S3 upload events
• Calls Amazon Rekognition to analyze the image
• Stores the detected labels and confidence scores in DynamoDB

View full Lambda code

---

3. Configuring the Event Trigger
I set up S3 to automatically trigger the Lambda function whenever a new image gets uploaded.

---

4. Setting Up IAM Permissions
The Lambda function needs specific permissions to:

• Read from the S3 bucket
• Analyze images with Rekognition
• Write results to DynamoDB For this project I added permissions for full access DynampDB, Rekognition and read on;y permissions for S3

---

5. Testing the Pipeline
When I upload an image:

• The trigger fires automatically
• Rekognition detects objects like "Person", "Car", "Building"
• Results get stored in DynamoDB with confidence scores

---

Tech Stack:

• AWS Lambda (Python 3.12) - Serverless compute
• Amazon S3 - Image storage and event source
• Amazon Rekognition - AI image analysis
• Amazon DynamoDB - NoSQL database for results

---

Performance Optimizations

Current Setup:

• Lambda timeout: 30 seconds (sufficient for most images)
• Max labels per image: 10 (configurable)

Possible Architecture Improvements:

• Separate Lambda function for DynamoDB writes to improve speed
• Add image resizing for faster processing
• Implement batch processing for multiple images

---

What I Learned

• How to build event-driven serverless architectures on AWS
• How to use Amazon Rekognition for automatic image analysis
• Why proper IAM permissions are crucial for Lambda functions
• That serverless scaling happens automatically without configuration

---

Personal Links: LinkedIn | GitHub