How Dengbêj Works
A fully serverless AI pipeline from news feeds to Kurdish radio.
RSS Feeds
News Sources
BBC, Deutsche Welle, Al Jazeera, and other international sources.
every 6 hours
Lambda
News Ingester
Fetches and parses RSS feeds, deduplicates articles, extracts metadata.
DynamoDB
Article Storage
Raw articles stored with source, timestamp, and feed metadata.
twice daily
Lambda + Bedrock
Today's 5 Curator
AI selects the five most important stories based on global significance and diversity.
DynamoDB
Briefing Storage
Curated briefings stored with date keys for daily retrieval.
15 min later
Lambda + Bedrock
Story Processor
Generates English summaries and Kurdish Kurmanji translations for each story.
Lambda + Bedrock
Program Generator
Classifies stories into regional programs and generates broadcast scripts.
Lambda + Bedrock + KurdishTTS / Polly
Daily Audio
Generates a narration script, synthesizes Kurmanji audio via KurdishTTS, and falls back to English via Amazon Polly.
S3
Audio Storage
MP3 files stored in S3 with public read access for streaming.
Lambda
News API
REST API serving today's briefing, programs, and audio URLs to the frontend.
Amplify
Frontend
Static HTML/JS app with bilingual UI, program selector, and audio player.
Pipeline Timing
Every 6 hours, fresh articles are ingested from international RSS feeds. Twice daily at 06:00 and 18:00 UTC, the AI curator selects the five most important stories for the day's briefing.
Fifteen minutes after curation, the story processor generates summaries and Kurdish translations. The program generator then classifies stories into regional programs and creates broadcast scripts. Finally, the audio pipeline synthesizes English narration and stores it for streaming.
The entire pipeline is orchestrated by Amazon EventBridge schedules, with each stage running as an independent Lambda function. No servers to manage, no infrastructure to maintain.