AWS Builder Center
Weekend Challange: Turn One Annoying Task Into an App : BuilderRadar-Ai

Weekend Challange: Turn One Annoying Task Into an App : BuilderRadar-Ai

A behind-the-scenes look at how i turned an annoying task into an app using Amazon EventBridge, AWS Lambda, DynamoDB, Strands Agents SDK, Amazon Bedrock AgentCore, and Amazon Nova Micro to automate hackathon discovery while keeping cloud costs low.

Series: Weekend Challange (2 articles)

  1. 2
    Weekend Challange: Turn One Annoying Task Into an App : BuilderRadar-Ai This article
AWS Cloud Club Member at PRMITR

Why I Built BuilderRadar AI

As a student passionate about cloud computing and building projects, I spend a lot of time looking for hackathons, cloud challenges, and opportunities that help me learn by doing. Most of these opportunities are available on platforms like Devpost, but finding the right one wasn't as simple as opening a website.
Every day, I found myself repeating the same routine: visiting Devpost, scrolling through new listings, reading descriptions, checking deadlines, and deciding whether a hackathon matched my interests in AWS, AI, or backend development. Some days there were no new opportunities at all, yet I had still spent valuable time searching.
After a while, I realized something.
The problem wasn't the lack of opportunities. The problem was the amount of time I was spending discovering them. As someone who enjoys building projects, I wanted to spend more time creating solutions than searching for them.
That simple frustration sparked an idea:
What if I could build a system that searched for opportunities on my behalf?
Not just a scraper that collected data, but an intelligent system that could identify new opportunities, avoid duplicates, understand what each opportunity was about, and present only meaningful results through a clean dashboard.
Another thought stayed with me throughout the project: cost.
As I was learning AWS and experimenting with AI services, I wanted to build something practical without worrying about unexpected cloud bills. That challenge influenced almost every architectural decision I made. Instead of keeping infrastructure running continuously, I designed an event-driven, serverless workflow where services execute only when they're needed. Instead of sending every dashboard request to an AI model, opportunities are analyzed once, stored, and reused.
The result became BuilderRadar AI: an event-driven opportunity discovery system that automates the repetitive work of finding hackathons while keeping the architecture simple, scalable, and cost-conscious.



The Problem :

Finding opportunities wasn't difficult. Finding the right opportunities consistently was.
As a student builder, I wanted to spend my time learning, building, and participating in hackathons, not repeatedly checking websites for updates. The manual process quickly became repetitive, especially when many listings were unchanged or unrelated to my interests.
I realized this wasn't a one-time task, it was a workflow that could be automated.
That led to a simple goal:
Build an event-driven system that automatically discovers new opportunities, filters out duplicates, enriches them using AI, and presents only meaningful results through a dashboard.



Designed Goals and Architecture Decisions :

Before choosing any AWS services, I defined a few design goals that would guide the architecture. Since BuilderRadar AI was built as a personal project, I wanted the solution to solve my problem without becoming expensive or difficult to maintain.
Rather than starting with technology, I started with a few simple questions:
  • How can I automate a repetitive task?
  • How can I keep cloud costs predictable?
  • How can I avoid processing the same opportunity multiple times?
  • How can I use AI only where it adds value?
These questions shaped every architectural decision that followed.
Design Decision
Why It Matters
Event-driven workflow
The pipeline runs only twice a day instead of continuously monitoring websites.
Serverless architecture
AWS Lambda executes only when invoked, eliminating idle compute costs.
SHA-256 deduplication
Duplicate opportunities are filtered before reaching the AI pipeline, avoiding unnecessary model invocations.
Limited scraping (2 opportunities per run)
Keeps database writes and AI processing predictable while capturing the newest listings.
Amazon Nova Micro
A lightweight model that is well suited for classification and summarization tasks while keeping inference costs low.
Model-free read path
The dashboard reads enriched data directly from DynamoDB without calling the AI model for every user request.
AWS Budget Alarm
Provides an additional safety net by alerting me if spending exceeds my expected usage.
These principles resulted in an event-driven pipeline where every component has a single responsibility. The scraper focuses on collecting data, the AI agent focuses on enriching it, and the frontend simply displays the processed results. This separation keeps the system easier to understand, maintain, and extend as new opportunity sources or AI capabilities are added.



Following an Opportunity Through the Pipeline

To better understand how BuilderRadar AI works, let's follow a single opportunity from the moment it appears on Devpost until it reaches the dashboard.
Step 1: Scheduling the Workflow
The pipeline begins with Amazon EventBridge, which triggers the scraping workflow twice a day. Instead of continuously running infrastructure or checking websites every few minutes, the system wakes up only at scheduled times, keeping compute usage and costs predictable.
Step 2: Collecting New Opportunities
Once triggered, two AWS Lambda functions fetch the latest opportunities from Devpost. Each scraper is intentionally limited to processing only the two newest listings per run. This small limit helps bound the number of database writes, AI invocations, and overall cloud costs.
For every opportunity, the scraper extracts essential information such as the title, description, deadline, source, and URL.
Step 3: Preventing Duplicate Processing
Before storing an opportunity, the system generates a SHA-256 hash using the normalized title and source.
This hash becomes the DynamoDB partition key.
Instead of performing a separate "Does this already exist?" query, the scraper performs a conditional write using attribute_not_exists(id). If the opportunity has already been processed, the write is rejected automatically. As a result, duplicate opportunities never reach the AI enrichment stage.
Step 4: Triggering AI Enrichment
Only when genuinely new opportunities are stored does the enrichment pipeline begin.
Rather than sending every scraped opportunity directly to an AI model, BuilderRadar AI stores the raw data first and then triggers a separate AI agent. This keeps data collection independent from AI processing and ensures that the model is only invoked when new information arrives.
The result is an event-driven workflow where repeated scraper executions don't repeatedly consume AI tokens for the same opportunity.



Architecture Flow :

Step
Service
Responsibility
1
Amazon EventBridge
Starts the workflow twice a day
2
AWS Lambda
Retrieves the latest opportunities
3
Amazon DynamoDB
Stores raw opportunities and prevents duplicates
4
Event Trigger
Starts AI enrichment only for new records



Adding Intelligence with an AI Agent :

Collecting opportunities was only half of the solution. The next challenge was understanding them.
A scraper can extract information such as the title, description, deadline, and URL, but it cannot determine whether an opportunity is actually relevant to someone interested in AWS, AI, or backend development. I wanted the system to do more than collect data. I wanted it to understand it.
To achieve this, I built an AI enrichment pipeline using the Strands Agents SDK, Amazon Bedrock AgentCore Runtime, and Amazon Nova Micro.
Instead of directly calling a language model from a Lambda function, I chose to use an AI agent. While a language model can generate text from a prompt, an agent can reason about a task and decide when to use external tools.
When a new opportunity is detected, the agent follows a simple workflow:
  1. Retrieve the raw opportunity using get_opportunity().
  1. Pass the opportunity to Amazon Nova Micro for analysis.
  1. Extract structured information such as the category, deadline, and a short relevance explanation.
  1. Save the enriched data back to DynamoDB using save_enrichment().
The Strands Agents SDK handles the interaction between the model and these tools. Instead of manually writing the logic for tool calling, response handling, and conversation flow, I simply define the available tools, and Strands manages the orchestration behind the scenes.
Although BuilderRadar AI currently performs only opportunity enrichment, this architecture makes it easy to extend the agent later with additional tools, such as sending notifications, integrating new opportunity sources, or ranking opportunities based on personal preferences.



AI Enrichment Workflow :

Why not call Nova Micro directly from Lambda?
For the current version of BuilderRadar AI, directly invoking Nova Micro from Lambda would have worked. I chose an AI agent because I wanted an architecture that could grow over time. By giving the agent access to tools, I can later add capabilities like querying additional databases, sending notifications, or integrating new opportunity sources without redesigning the workflow. It was a decision made with future scalability in mind rather than solving only today's problem.


Designing for Cost and Efficiency :

One of the biggest constraints while building BuilderRadar AI wasn't the technology. It was cost.
As a student experimenting with AWS services and generative AI, I wanted to learn by building without worrying about unexpected cloud bills. Rather than optimizing costs after the project was finished, I made cost-conscious decisions throughout the design process.
The goal was simple: only spend money when the system is doing useful work.
This philosophy influenced every stage of the architecture.




Challenges and Key Learnings :

Every project teaches lessons that you don't anticipate at the beginning, and BuilderRadar AI was no exception. While building it, I discovered that designing the architecture was often more challenging than writing the code.
A few lessons stood out:
  • Design for cost from the beginning. As a student builder, cloud costs were always on my mind. This encouraged me to adopt an event-driven, serverless architecture where services only run when necessary.
  • Separate data collection from AI processing. Storing raw opportunities before AI enrichment made the pipeline easier to maintain, prevented duplicate AI calls, and made the system more resilient.
  • Choose the right model for the job. Not every AI task requires a large model. Amazon Nova Micro was more than capable of classification and summarization while keeping inference costs low.
  • Think beyond today's requirements. Using an AI agent with tools gave me a flexible architecture that can easily grow to support new data sources, notifications, or additional automation in the future.
More than anything, this project changed the way I think about cloud architecture. Instead of asking "Which AWS service should I use?", I now start by asking "What problem am I solving, and what is the simplest, most efficient way to solve it?"

Conclusion

BuilderRadar AI started as a simple idea to save time searching for hackathons, but it became much more than that. It was an opportunity to explore serverless architecture, AI agents, event-driven design, and cost-conscious engineering on AWS.
The biggest takeaway wasn't learning a new AWS service. It was understanding that good architecture is built on thoughtful decisions. Every service in this project exists for a reason, whether it's reducing cost, improving scalability, or keeping the system simple to maintain.
This project is still evolving, and I look forward to adding more opportunity sources, smarter personalization, and additional AI capabilities in future iterations.
If you're learning AWS or exploring AI agents, I hope BuilderRadar AI inspires you to solve a real problem that matters to you. Sometimes the best projects don't start with a technology, they start with a small frustration that you're curious enough to automate.
Resources







Series: Weekend Challange (2 articles)

  1. 2
    Weekend Challange: Turn One Annoying Task Into an App : BuilderRadar-Ai This article
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article