
Weekend Creative Challenge: PromptPilot Studio
PromptPilot Studio is a lightweight creative AI tool that transforms rough ideas into structured creative briefs using Amazon Bedrock and Amazon Nova Lite. Built during the AWS Weekend Creative Challenge, this project explores prompt engineering, structured AI output, and a simple serverless-ready architecture.
Weekend Creative Challenge: PromptPilot Studio
For this weekend's challenge, I wanted to build something that solves the dreaded "blank page" problem. Whether you're a marketer, a creator, or just someone with a rough idea, getting from a single sentence to a cohesive creative brief is tough. That's why I built PromptPilot Studio—a lightweight web application that takes a simple, unrefined idea and uses Amazon Bedrock to expand it into a comprehensive creative package.
The Idea Behind PromptPilot Studio
The vision was simple: I wanted an application that felt like a creative director in your browser. Instead of just returning a wall of generic AI text, the tool needed to enforce structure. I wanted the output to be predictable, visually clean, and immediately actionable for creative expression.
What PromptPilot Studio Does
When you open PromptPilot Studio, you are greeted with a clean, editorial-style interface. You type in a rough idea (e.g., "Students should stop wasting food in college cafeterias"), select your target audience, choose a tone like "Emotional" or "Bold," and specify a creative format like a Campaign or a Product Concept.
When you hit "Transform Idea", the frontend sends this data to the backend, which prompts Amazon Nova Lite. The AI doesn't just chat back; it returns a strict JSON object containing:
- A memorable creative concept name
- A short, catchy tagline
- A 100-180 word narrative story bringing the concept to life
- Visual direction (mood, colors, typography, imagery, and composition)
- A social media caption
- An unexpected "creative twist" to make the concept stand out
The UI parses this and renders a beautiful, structured brief that you can copy or download.
How I Built It
I approached this with a focus on simplicity and reliability.
Frontend: I stuck to vanilla HTML, CSS, and JavaScript. No heavy frameworks, no build steps. I wanted a responsive, fast-loading UI with dark/light mode support. I added strict client-side validation to ensure inputs aren't too long, and implemented an
AbortController to gracefully handle network timeouts.Backend: I wrote a Python Lambda-style handler (
lambda_function.py ). Its primary job is to take the JSON payload from the frontend, construct a highly specific system prompt, and use boto3 to invoke Amazon Bedrock.Prompt Engineering: To ensure the output was always usable, I engineered the prompt to demand strict JSON formatting matching a specific shape. I also wrote regex fallback logic in Python to extract the JSON payload just in case the LLM wrapped its response in markdown code fences.
Local Testing: To iterate quickly without deploying to AWS every five minutes, I wrote a
local_server.py script. It uses Python's built-in BaseHTTPRequestHandler to mimic how an API Gateway would invoke the Lambda function.The AWS Architecture

Building this wasn't perfectly smooth. I hit some very real roadblocks:
- Inference Profile Confusion: Initially, I tried invoking the
amazon.nova-lite-v1:0model ID directly. Bedrock threw aValidationException: Operation not allowed. After some digging, I learned that for Nova models, I had to route the request through a cross-region inference profile (us.amazon.nova-lite-v1:0). I updated my environment variables and the code worked perfectly.
- Missing Local Credentials: My local development machine didn't have valid AWS credentials configured. When my local server tried to invoke Bedrock,
boto3threw aNoCredentialsError. I couldn't perform a true end-to-end test on my local machine.
- The Deployment Blocker: To bypass my local credential issues, my plan was to log into the AWS Console, open AWS CloudShell, and use AWS SAM to deploy my prepared
template.yaml. Unfortunately, because my AWS account is brand new, it was stuck in an "account verification in progress" state. This completely blocked me from opening CloudShell and executing the deployment.
What I Learned
This weekend taught me a lot about building resilient apps around Generative AI:
- Amazon Bedrock Quotas: I learned firsthand how Amazon manages capacity for newer models using cross-region inference profiles, and how to configure
boto3to target them correctly.
- Defensive API Design: Because LLM outputs can be unpredictable, I learned how crucial it is to write robust backend parsing logic. Anticipating markdown fences and using regex to extract JSON saved my application from crashing.
- Secure Configuration: I practiced strict credential hygiene. I ensured no AWS keys were ever hardcoded in the JavaScript or Python, utilized
.envfiles locally, added them to.gitignore, and designed the SAM template to inject environment variables securely into Lambda.
- Least Privilege IAM: In writing my SAM template, I learned how to scope Lambda execution roles down to exactly what was needed—specifically just the
bedrock:InvokeModelaction for the exact Nova Lite ARN—rather than relying on broad administrator access.
What’s Next
The code is written, the SAM templates are prepared, and the deployment runbook is documented in the repository. The immediate next steps are:
- Wait for AWS account verification to finish so CloudShell becomes available.
- Run
sam buildandsam deployto provision the API Gateway and Lambda.
- Inject the live API URL into the frontend and upload it to an S3 bucket for static hosting.
- Perform a real end-to-end generation test in the browser.
Project Links
GitHub Repository: https://github.com/Ayush-0918/promptpilot-studio
(Live deployment is still pending AWS account verification.)
Building PromptPilot Studio over the weekend was a fantastic exercise in architecture and prompt engineering. Even though I hit a deployment wall at the very end, working through the Bedrock API and designing a system that forces AI into a structured, usable format was incredibly rewarding.
#creative-expression #challenge
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article