Inspiration
Film productions can lose $150K–$250K from a single day of delay on a $5M production. CineOps asks: What if AI could turn a producer's free-text crisis into a coordinated response in under 90 seconds?
What it does
CineOps is a multi-agent AI production command center. One event-like “Actor unavailable tomorrow”-activates Planner, Scheduling, Budget, Legal, and Notification agents in parallel.
How we built it
A FastAPI + React/TypeScript architecture orchestrates five concurrent agents using Google ADK, Gemini 2.5 Flash, MCP, and live BigQuery data. The platform runs in Docker and deploys to Google Cloud Run.
Challenges we ran into
Async integration with the synchronous Gemini SDK initially caused event-loop bottlenecks. Gemini's free-tier quota limit also forced us to build a three-tier fallback system that keeps CineOps operational even without credentials or quota.
Accomplishments that we're proud of
CineOps delivers a plan under virtually any infrastructure condition through ADK → Gemini → rule-based fallback. A 49-test offline suite validates agents, MCP, routing, and audit workflows without requiring GCP credentials.
What we learned
MCP showed us how much stronger agent reasoning becomes when grounded in live production data instead of assumptions. We also learned that async correctness is critical when multiple agents execute concurrently.
What's next for CineOps -Film Production Command Center
Our BigQuery audit trail creates a valuable feedback signal for improving routing and decision quality over time. Next, we're adding real-time streaming via Server-Sent Events, so producers see agent results as soon as they arrive.
