GLC Newsroom by Ndotonic

Inspiration

I come from a news reporting background, so I know how much work it takes to keep a beat current, verify sources, and get a bulletin on air. Independent broadcasters and creator-led media teams cannot staff a newsroom around the clock. Important stories move between scheduled bulletins, while manual web research is slow, repetitive, and difficult to verify. I built GLC News to continuously watch selected beats and turn trustworthy developments into broadcast-ready audio.

What it does

GLC News is an agentic audio-news production workflow for GLC Media. An editor enters a topic such as world news, business, health, technology, or sports. Parallel Search retrieves current source URLs and excerpts as a constrained research packet. Gemini writes a concise desk-specific reporter script from those live facts, and Google Cloud Text-to-Speech renders the final professional audio bulletin.

GLC News now also includes updates monitoring with Parallel Monitor. An administrator creates one recurring monitor for each newsroom beat and chooses an hourly or daily schedule. When Parallel detects a material development, it sends a signed webhook to GLC News.

The application verifies the webhook HMAC signature and timestamp, maps the event to its topic and desk, retrieves the complete Parallel event group, stores its summary and source citations in a newsroom inbox, and removes duplicate events. The next manual bulletin merges those monitored developments with a fresh Parallel Search before Gemini writes its script.

Administrators can create, update, trigger, list, and permanently cancel monitors. Active monitors are capped to control cost. Automatic bulletin generation remains gated so editors retain control over publication.

How we built it

• Python and Flask provide the application and authenticated newsroom APIs.
• PostgreSQL stores recurring monitors, detected events, citations, task state, and bulletin metadata.
• Parallel Search supplies current source-backed research at generation time.
• Parallel Monitor watches long-lived news beats and pushes material changes through signed webhooks.
• Gemini writes desk-specific scripts from constrained research packets.
• Google ADK defines the specialized newsroom agent roles.
• Google Cloud Text-to-Speech renders the broadcast audio.
• GLC Media distributes finished bulletins through its existing radio infrastructure.

Challenges we faced

Parallel's stable Monitor API differs from its earlier alpha interface, so we implemented the current /v1 create, update, trigger, event retrieval, and irreversible cancellation lifecycle. We also implemented Standard Webhooks HMAC verification against the exact raw request body, replay-age checks, idempotent event storage, and graceful fallbacks that do not break existing audio generation.

Another challenge was separating two different research jobs: Monitor answers "what changed since the last run," while Search answers "what is true right now." Combining both without duplicating facts required a persistent newsroom inbox and a deterministic merge before Gemini writes each report.

Real audio generation can take several minutes because the system performs research, script generation, and speech rendering. The demonstration therefore shows the real input and completed 2:07 bulletin while compressing the waiting period rather than suggesting instant generation.

Accomplishments that we're proud of

• A real end-to-end audio newsroom rather than a standalone research chatbot.
• Continuous beat monitoring with source citations and secure webhook handling.
• Fresh Parallel Search research for every manually generated bulletin.
• Desk-specific Gemini scripts and production Google Cloud TTS audio.
• Safe degradation when an external AI or research service is unavailable.

What we learned

Agentic media systems need more than generation. They need event memory, source provenance, deterministic routing, security boundaries, human approval, and reliable fallbacks. Parallel is valuable both as the live research layer and as the signal that tells the newsroom when a beat is worth revisiting.

What's next

Next we will expose the stored event inbox in the newsroom dashboard, deploy the orchestration through Google Cloud's managed agent platform, add editor-approved automatic audio bulletins for breaking stories, and extend the verified workflow to multilingual radio reports.

Built With

• agentic-ai
• docker
• flask
• gemini
• google-adk
• google-cloud
• google-cloud-text-to-speech
• hmac
• parallel-monitor-api
• parallel-search-api
• postgresql
• python
• rest-api
• webhooks