ECHO-CHAMBER

Inspiration

Screenwriters often write dialogue in a vacuum, resulting in outdated tropes or shallow subcultural caricatures. We wanted to build an agent that grounds narrative dialogue in real-time, live web discourse without stripping away creative control.

What it does

ECHO-CHAMBER is an autonomous agent pipeline built on Google Cloud Agent Builder and Gemini 2.5. It ingests script drafts, isolates regional or subcultural dialogue markers, queries live web discourse via the Parallel Search & Extract MCP server, and rewrites scenes to eliminate outdated tropes—producing an updated script alongside a transparent "Cultural Authenticity Ledger."

How we built it

• Orchestration: Google Cloud Agent Builder & Gemini 2.5 Pro for multi-step reasoning.

• Grounding & Web Extraction: Parallel Search & Extract MCP Server to fetch unstructured subcultural web data and live vernacular.

• Backend & Execution: Python serverless environment running the Google GenAI SDK to chain agent steps together.

Challenges we ran into

Balancing real-world slang insertion with script narrative flow—ensuring the agent rewrites dialogue so it feels natural, grounded, and contextually accurate rather than over-satirized.

Accomplishments that we're proud of

Successfully bridging generative scriptwriting with dynamic web search via Parallel MCP, creating a verifiable subcultural authenticity pipeline in seconds.

What we learned

How to effectively pass unstructured MCP tool outputs into Gemini's multi-step reasoning loops to produce structured outputs like our Cultural Authenticity Ledger.

What's next for ECHO-CHAMBER

• Multi-Agent Writer's Room: Expanding the pipeline from a single dialogue rewriter into a multi-agent orchestration setup (e.g., assigning separate Gemini agents to handle Character Backstory Consistency, Dialect Accuracy, and Narrative Pacing concurrently).
• Live Video & Voice Table Reads: Integrating Gemini 2.5 multimodal audio capabilities to auto-generate vocal table reads of rewritten scenes, allowing directors to listen to natural dialect stress and cadence in real time.
• Expanded MCP Grounding Ecosystem: Connecting additional MCP servers (such as ClickHouse telemetry for streaming viewer sentiment) to cross-reference subcultural accuracy with live audience engagement metrics.

Built With

• gemini
• google-cloud-agent-builder
• javascript
• parallel-mcp
• python