
Weekend Creative Agent Challenge: Dengbêj AI — Bêje, an Always-On Kurdish News Radio
Dengbêj AI evolves from a Kurdish storytelling prototype into Bêje, an always-on news-radio agent that automatically collects current stories, generates Kurdish content with Amazon Bedrock, and prepares new programs before listeners return.
Earlier this year, I introduced Dengbêj AI as part of the AWS 10,000 AI Ideas Challenge.
The original vision was to explore how generative AI could support Kurdish storytelling and make global information more accessible to Kurdish speakers.
That idea was selected as a semi-finalist.
Later, an AWS Weekend Creative Challenge motivated me to return to the unfinished project. I completed its serverless news pipeline and published the first working prototype.
This is Volume 3 of that journey.
From an App to an Always-On Agent
The previous version of Dengbêj AI could collect news, generate Kurdish content, and present stories through a radio-inspired interface.
However, the most important feature was happening behind the interface.
The system did not need to wait for someone to open the application and enter a prompt. It could wake up automatically, discover new stories, generate new programs, and have them ready when the listener returned.
That makes Bêje an always-on content agent.
Its cycle is:
Observe → Select → Generate → Store → Publish → RepeatAmazon EventBridge wakes the agent on a schedule. AWS Lambda collects stories from selected RSS sources and organizes them for processing. Amazon Bedrock generates Kurdish headlines, summaries, and radio-style scripts.
The completed programs are stored in Amazon DynamoDB, while generated audio is stored in Amazon S3. AWS Amplify hosts the public application.
When someone opens Bêje, the content has already been prepared.
The Bêje! / Tell me! button does not start the complete generation process. It retrieves and plays a program that the agent created earlier.
Why I Froze the Previous Version
Once the previous challenge version was working reliably, I had to decide whether to continue changing it or preserve it.
I decided to freeze it.
That version represented the stage documented in my second article. Continuing to redesign the same deployment could have broken the experience for readers opening the original link.
I therefore kept the stable version on its own branch and created a separate branch and AWS Amplify deployment for Bêje V2.
Freezing it did not mean that development was finished.
It meant that one stage was stable enough to preserve while the next stage began.
Introducing Bêje
The new version is centered around one Kurdish word:
Bêje.
It means “Tell me.”
I wanted the application to feel less like a traditional news feed and more like asking someone to tell you what is happening.
Listeners can choose an available topic or region and press Bêje! The English interface presents the same invitation as Tell me!
V2 also adds:
- A bilingual Kurdish and English interface
- Kurdish headlines and summaries
- Topic-based and regional programs
- Continuous audio playback
- Previous and next controls
- Better mobile navigation
- Clear unavailable, loading, and error states
- About and How It Works pages
- Improved accessibility
- Correct use of the Kurdish spelling Dengbêj
The goal was not to add as many features as possible. It was to make the existing system clearer, more reliable, and easier to understand.
AWS Architecture
The agent uses:
- Amazon EventBridge to activate scheduled workflows
- AWS Lambda for ingestion, curation, processing, APIs, and audio preparation
- Amazon Bedrock for Kurdish headlines, summaries, and program scripts
- Amazon DynamoDB for articles, programs, and completed briefings
- Amazon Polly for the current English narration
- Amazon S3 for generated audio
- AWS Amplify for public hosting and branch deployments
The pipeline also uses content fingerprints. If the selected stories have not changed, the agent avoids generating the same program again. This reduces unnecessary model requests and helps control operating costs.
What I Learned
This stage taught me that an agent does not have to be a chatbot.
An agent can work quietly in the background, follow a schedule, observe changing information, make processing decisions, create something useful, and prepare it before the user arrives.
I also learned that building new features is only one part of development. Stable branches, tests, deployments, rollback options, and honest documentation are equally important.
AWS Builder Challenges continue to motivate me to return to my projects and turn unfinished ideas into public releases.
My participation has already earned me three AWS Builder jackets and AWS credits. Those rewards are encouraging, but the experience has been even more valuable. With every challenge, I learn more about AWS architecture, generative AI, automation, testing, deployment, and product development.
Dengbêj AI did not develop in a straight line. It paused, returned, changed direction, and became stronger through that process.
What Comes Next
The biggest missing feature is native Kurdish speech.
The application provides Kurdish text, but the current audio uses English narration because Amazon Polly does not offer the Kurdish voice required by the project.
A future version could integrate dedicated Kurdish speech technology and provide both Kurmanji and Sorani options.
I also want to explore more Kurdish news sources, improved translation review, stronger pipeline monitoring, community contributions, local journalism, and oral-history collections.
The first article introduced the vision.
The second turned that vision into a working prototype.
This third volume transforms it into a clearer always-on experience.
Dengbêj is the tradition behind the project.
Bêje is the invitation to listen.
Explore the Project
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article