AI Production Intelligence

Inspiration

Film and audiovisual productions generate a constant stream of operational events: weather risks, equipment failures, actor delays, and schedule disruptions. These events are often handled separately, making it difficult to see their combined impact on production continuity.

We built AI Production Intelligence to give production teams an AI-powered operational view of what is happening and, more importantly, what they should do next.

What it does

AI Production Intelligence analyzes real production events stored in ClickHouse and uses Google Gemini to generate a concise operational assessment and recommendation.

The system identifies the most relevant production risk, considers its impact on continuity and scheduling, and recommends practical next actions.

For example, when a production has an outdoor scene threatened by heavy rain, while also dealing with an actor delay and a previous camera failure, the AI can identify the weather risk as the immediate priority and recommend moving the shoot indoors or rescheduling while considering actor availability.

How we built it

The application is built with Java and Spring Boot.

Production events are stored and queried in ClickHouse. Google Gemini running through Vertex AI provides the intelligence layer that analyzes the production context.

We also integrated Google ADK and ClickHouse MCP to enable an agent-based architecture where the AI agent can retrieve production information through MCP tools.

The backend is containerized with Docker and deployed on Render, with a web frontend providing the production intelligence dashboard.

Why ClickHouse

ClickHouse is the foundation of the production event data layer. Production events can be queried efficiently and provided as context for AI analysis.

This makes it possible to move from raw operational events to an actionable production recommendation.

Challenges

One of the biggest challenges was integrating an agentic workflow with a remote MCP server while maintaining a reliable production experience.

We implemented a resilient fallback flow so that if the ADK + MCP connection encounters a runtime problem, the application can still retrieve the production data from ClickHouse and generate the analysis with Gemini.

This allowed us to maintain a functional production system while continuing to explore the agentic MCP architecture.

What we learned

We learned that building an AI agent for production environments is not only about generating good answers. Reliability, data grounding, observability, and graceful failure are equally important.

The project also gave us practical experience integrating ClickHouse, MCP, Google ADK, Gemini, Vertex AI, Spring Boot, and a real production-style workflow.

What's next

Future versions could incorporate richer production schedules, dependency analysis between events, predictive risk detection, automated notifications, and additional production-management tools exposed through MCP.

Our goal is to evolve AI Production Intelligence from an analysis dashboard into an operational AI assistant for audiovisual production teams.

Built With

• adk
• ai
• angular.js
• api
• boot
• clickhouse
• docker
• gemini
• google
• java
• maven
• mcp
• render
• rest
• spring
• typescript
• vertex