studioops-agentic-ai

StudioOps AI

Agentic Production Command Center for Film & Television (Agentic Cinema: The Blockbuster Hackathon - Grafana Track)

Overview

StudioOps AI is a web-based AI incident-command system designed to resolve production disruptions (e.g. weather shutting down an exterior shoot, actor unavailability, schedule slippages).

The application utilizes Google ADK and Gemini 2.5 Flash on Vertex AI to analyze production data retrieved live via Grafana MCP. Multiple specialist agents run concurrently to assess schedule, budget, and risk impacts. A Recovery Planner agent synthesizes these findings into 3 ranked options with transparent trade-offs. To enforce strict human governance, the final plan awaits the Producer's explicit approval before any action is logged.

Screenshots

Command Center and Live Operations

StudioOps AI Command Center: an incident commander provides operational context while the Coordinator, Schedule, Budget, Risk, and Recovery Planner agents await or execute analysis. Grafana MCP provides an auditable connection to live operational events.

AI Recovery Plan with Human Approval

Human-governed AI decision support: StudioOps evaluates multiple recovery paths, recommends a transparent option, and requires an authorized producer to approve or revise the plan before operational action is taken.

Workflow & Architecture

Operational lifecycle: Incident input → specialist agent analysis → recovery options and trade-offs → human approval or revision → audit event and Grafana MCP visibility.

graph TD;
    Incident[Incident Trigger] --> Coordinator[Coordinator Agent];

    subgraph GoogleADK ["Google ADK & Vertex AI (Gemini 2.5 Flash)"]
        Coordinator
        
        subgraph Specialists [Parallel Specialist Analysis]
            Schedule[Schedule Agent]
            Budget[Budget Agent]
            Risk[Risk Agent]
        end
        
        Coordinator --> Specialists
        Specialists --> Recovery[Recovery Planner Agent]
    end

    Coordinator -->|Grafana MCP Tool| Loki[Grafana Loki Logs / MCP];
    Loki --> Coordinator;

    Recovery --> Human[Human Approval Gate];
    Human -->|Approve Plan / Reject| Audit[Audit Event & Grafana MCP Logging];

Agent Roles

1. Production Coordinator Agent: Receives incident details, queries live operational context and Loki logs via Grafana MCP, and delegates tasks to specialists.
2. Schedule Agent: Evaluates timing constraints, permits, and actor availability to calculate schedule delays.
3. Budget Agent: Estimates overtime exposure, delay costs, and financial impact.
4. Production Risk Agent: Evaluates safety, equipment exposure, and assigns a 0–100 operational risk score.
5. Recovery Planner Agent: Synthesizes specialist findings into 3 ranked recovery options with quantified costs, delays, risks, advantages, and drawbacks.

Human Governance

StudioOps enforces strict human-in-the-loop safeguards at every step:

• Decision Support, Not Autonomous Action: AI agents only provide structured analysis and options; they never execute irreversible production changes autonomously.
• Quantified Trade-offs: Every recommended option presents explicit metrics for incremental cost, delay in minutes, operational risk score (0–100), advantages, and drawbacks.
• Producer Approval Gate: A human producer must explicitly review and select Approve Plan or Reject / Revise.
• Audit-Ready Logging: Approved or rejected decisions are recorded as structured audit events.
• Transparent Operational Visibility: Grafana MCP provides live operational-event visibility; any backend tracking issue or fallback is reported transparently rather than hidden.

Google Cloud Technologies

• Google ADK (Agent Development Kit): Powers the Agent, ParallelAgent, and SequentialAgent orchestrations.
• Google GenAI / Gemini: Utilizes gemini-3.6-flash for high-speed deterministic generation.

Grafana MCP Integration

The application explicitly leverages the Grafana MCP Server via standard Stdio to connect to Grafana Cloud Loki.

• Tools used: query_loki_logs
• Data flow: The Coordinator Agent queries Loki for recent operational events. If the live credentials are not set, the app gracefully falls back to a clearly-labeled "DEMO MODE".

Local Setup

1. Create Virtual Environment

python -m venv venv
# Windows
venv\Scripts\activate
# Mac/Linux
source venv/bin/activate

2. Install Dependencies

pip install -r requirements.txt

3. Environment Variables

Copy .env.example to .env and fill it out:

cp .env.example .env

Add your GOOGLE_API_KEY (or vertex equivalents). To test Grafana Live MCP, provide GRAFANA_URL and GRAFANA_SERVICE_ACCOUNT_TOKEN.

4. Run Application

uvicorn app.main:app --reload

Hackathon Demo Flow

1. Load the web interface.
2. Observe the UI is in DEMO mode if no Grafana credentials are provided.
3. Trigger the "Severe weather halts exterior shoot" incident.
4. Watch the AI Crew status dynamically update.
5. Review the final structured recovery plan.
6. Click "Approve Plan".

Future Enhancements

• Grounded production knowledge retrieval: Add a Google Cloud–native RAG capability to retrieve relevant call sheets, permit constraints, safety procedures, crew agreements, and location requirements before the agents produce recovery options.
• Persistent decision memory: Store approved and rejected recovery plans so StudioOps can surface comparable past disruptions, explain precedent, and improve consistency across future incidents.
• Cited recommendations: Show the operational logs, production documents, and prior decisions that informed each recovery option, while preserving the Producer approval gate.
• Production-grade RAG rollout: Extend StudioOps after the core Grafana MCP workflow is established, using a curated production-document corpus, tested retrieval quality, source citations, and an updated end-to-end demo. This preserves Grafana MCP as the live operational-context integration rather than treating RAG as a substitute for it.

License

MIT License