Inspiration

As a first-year engineering student, I was always curious about artificial intelligence and how real-world systems actually work behind the scenes. Getting selected for the AWS AI & ML Scholarship gave me the perfect platform to dive in.

The idea for this project came from one of our capstone challenges for a fictional logistics company, Scones Unlimited. I was fascinated by how a simple image classification task—detecting whether a delivery agent was using a bicycle or motorcycle—could lead to smarter routing, faster deliveries, and real business impact. It was the perfect blend of AI, automation, and cloud scalability.

What it does

This project classifies delivery vehicle images (bike vs motorcycle) in real-time using a trained CNN model hosted on Amazon SageMaker. The entire pipeline is automated using AWS Step Functions, with Lambda functions handling image preprocessing, inference, and confidence checking.

The model helps optimize logistics by assigning deliveries based on the type of vehicle—making operations smarter, faster, and more efficient.

I also built a no-code ML model using SageMaker Canvas to classify different types of flowers, giving me a broader understanding of ML workflows with and without code.

How we built it

I used the following services to build a fully functional ML workflow:

Amazon S3 to store images and datasets

Amazon SageMaker Studio to train a CNN image classifier using transfer learning

AWS Lambda to preprocess images, call the SageMaker endpoint, and validate results

AWS Step Functions to automate and orchestrate the end-to-end workflow

Amazon CloudWatch to track errors and monitor performance

SageMaker Canvas (AutoML) to experiment with no-code model building for flower classification

Here’s how the workflow runs:

An image is uploaded to S3

Step Functions trigger a Lambda to serialize the image

The second Lambda sends the image to the SageMaker model

The third Lambda checks if the prediction confidence meets the threshold

Based on the result, it’s either logged or flagged for review

The no-code Canvas workflow was equally intuitive:

Selected "Image Classification" as the problem type

Uploaded my flower dataset

Let Canvas handle preprocessing and model training

Evaluated the model using test images

Received clear performance metrics—without writing any code!

Challenges we ran into

Working with base64-encoded images in Lambda was tricky due to payload size limits

Chaining Lambda functions using Step Functions required deep understanding of event payloads and transitions

Confidence threshold logic had to be precise to avoid false positives

Balancing inference accuracy and response time was a key deployment concern

It was my first time deploying a model, so learning IAM roles and resource permissions took time but paid off

Accomplishments that we're proud of

Built and deployed a real-time, serverless ML pipeline completely on AWS

Achieved 94% model accuracy with fast and reliable inference

Used Step Functions to make the workflow production-ready with retries and error handling

Created a no-code ML solution in SageMaker Canvas and understood how AutoML works internally

Learned to think beyond just code—to system design, cloud architecture, and scalability

What we learned

How to go from model training to deployment in a real-world production environment

The value of serverless tools like Lambda and Step Functions for automation

The importance of monitoring, error-handling, and cloud cost-awareness

That AutoML tools like Canvas are powerful for both beginners and rapid prototyping

That it's entirely possible to build scalable AI solutions

What's next for MY AWS AI/ML JOURNEY AS A FIRST-YEAR STUDENT

This project has only deepened my passion for machine learning and cloud engineering. Moving forward, I plan to:

Explore real-time object detection and deploy models using Amazon SageMaker Edge

Dive into natural language processing for sentiment and bias detection

Learn more about AWS Bedrock, Amazon Rekognition, and Generative AI tools

Contribute to open-source projects and help others in the AWS community

Build a portfolio of deployable AI applications to solve problems in education, healthcare, and sustainability

Built With

Share this project:

Updates