How I Built a Serverless AI Image Analyzer in 40 Minutes (Free Tier)

Stop memorizing. Start building. Here is how I connected S3, Lambda, and Rekognition to build a real-world project.

The Problem with "Tutorial Hell"

I’ve spent weeks reading about S3 Buckets, IAM Roles, and Lambda triggers. But reading isn't knowing. Building is knowing.

Tonight, I challenged myself: Can I build a fully serverless AI image analyzer in under 60 minutes?
It took me around 40 minutes.

The Architecture

We are building an event-driven workflow. No servers to manage.

Trigger: You upload an image to an S3 Bucket.

Compute: This event triggers an AWS Lambda function (Python).

Intelligence: The code sends the image to Amazon Rekognition (AI).

Output: The AI tells us what is in the image via CloudWatch Logs.

Step 1: The Input (S3)

First, we need a place to drop our files.

Go to S3 in the AWS Console.

Click Create bucket.

Name it something unique (e.g., image-analysis-lab-yourname).

Leave all settings as default (Block Public Access should be ON).

Click Create bucket.

Step 2: The Security (IAM)

How I Built a Serverless AI Image Analyzer in 40 Minutes (Free Tier)

Subtitle: Stop memorizing. Start building. Here is how I connected S3, Lambda, and Rekognition to build a real-world project.
The Problem with "Tutorial Hell"

I’ve spent weeks reading about S3 Buckets, IAM Roles, and Lambda triggers. But reading isn't knowing. Building is knowing.

Tonight, I challenged myself: Can I build a fully serverless AI image analyzer in under 60 minutes? It took me around 40 minutes.

Here is the step-by-step guide on how you can build this too during your lunch break.

The Architecture

We are building an event-driven workflow. No servers to manage.

Trigger: You upload an image to an S3 Bucket.

Compute: This event triggers an AWS Lambda function (Python).

Intelligence: The code sends the image to Amazon Rekognition (AI).

Output: The AI tells us what is in the image via CloudWatch Logs.

Step 1: The Input (S3)

First, we need a place to drop our files.

Go to S3 in the AWS Console.

Click Create bucket.

Name it something unique (e.g., image-analysis-lab-yourname).

Leave all settings as default (Block Public Access should be ON).

Click Create bucket.

Step 2: The Security (IAM)

This is where many beginners get stuck. Our Lambda function needs an ID badge (Role) to access the photos and talk to the AI.

1.Go to IAM -> Roles -> Create role.

2.Select AWS Service and choose Lambda.

3.Add these three permissions:

• AmazonS3ReadOnlyAccess (To see the file)

• AmazonRekognitionReadOnlyAccess (To ask the AI)

• AWSLambdaBasicExecutionRole (To write logs)

4.Name the role "Lambda-AI-Role" and create it.

Step 3: The Logic (Lambda)

Now, the brain of the operation.

1. Go to Lambda -> ** Create function.**

2.Select Python 3.12 as the runtime.

Under Permissions, choose "Use an existing role" and select your "Lambda-AI-Role."

4.Create the function.

Once created, scroll up to the Function Overview and click + Add trigger.
Select S3, choose your bucket, and save. Now your code wakes up whenever a file lands in the bucket.

Step 4: The Code (And the Bug I Found)

This was my "Aha!" moment.

My first attempt failed with an "InvalidS3ObjectException."

I uploaded a file named "Test Image 1.jpg". The issue: S3 automatically encodes spaces in filenames to "+" (so it became "Test+Image+1.jpg").
My code was looking for the file with "+" in the name, which didn't exist on the disk.

The Fix: I had to import "urllib.parse" to decode the filename back to normal text.

Here is the full, working code:

import json
import boto3
import urllib.parse

def lambda_handler(event, context):
s3_client = boto3.client('s3')
rekognition_client = boto3.client('rekognition')

# Get the bucket name and filename from the event
bucket_name = event['Records'][0]['s3']['bucket']['name']

# DECODE THE FILENAME (The crucial fix!)
raw_file_name = event['Records'][0]['s3']['object']['key']
file_name = urllib.parse.unquote_plus(raw_file_name)

print(f"Analyzing: {file_name} from {bucket_name}")

try:
    # Call Amazon Rekognition
    response = rekognition_client.detect_labels(
        Image={'S3Object': {'Bucket': bucket_name, 'Name': file_name}},
        MaxLabels=10,
        MinConfidence=75
    )

    # Print results to logs
    print("--- RESULTS ---")
    label_names = []
    for label in response['Labels']:
        print(f"Found: {label['Name']} ({label['Confidence']:.2f}%)")
        label_names.append(label['Name'])

    return {
        'statusCode': 200,
        'body': json.dumps(f"Found: {', '.join(label_names)}")
    }

except Exception as e:
    print(f"ERROR: {str(e)}")
    raise e

Step 5: The Result

I uploaded an image of my desk setup. Checking the CloudWatch Logs, I got this:

Found: Computer (100.00%)
Found: Laptop (100.00%)
Found: Hardware (90.87%)

It works. Zero servers provisioned. Total cost: $0.00 (Free Tier).

Conclusion

Don't just collect certifications. Collect projects. The "magic" of AI is accessible to anyone willing to connect the dots and debug the errors.

Go build this!

Comments

[Aryan Choudhary]

I've been meaning to get into working with AWS for ages, but it's always seemed like a steep learning curve.
Your explanation really helped break it down in a way that feels approachable though, so thanks for sharing your knowledge in a way that's easy to follow.
Your post feels like a great starting point for me Ali.

[Ali-Funk]

This first comment made my day, Aryan!

The AWS learning curve can look like a „wall“, but once you start breaking it into small bricks (like Lambda and S3), it becomes a staircase.

If you decide to give this project a try and get stuck anywhere, feel free to drop a question here. Building is the best way to learn.

Good luck starting