Cinema Sentinel — AI Stream Health Agent

🎬 Cinema Sentinel — Autonomous Video Streaming Health Agent

Problem

Video streaming platforms lose millions to quality degradations that go undetected for hours. Buffering spikes, bitrate drops, and CDN failures are detected manually through user complaints — too slow for live media.

Solution

Cinema Sentinel is an autonomous AI agent built on Google ADK + Gemini 3.5 Flash that monitors video streaming infrastructure in real-time. It uses Grafana MCP to query observability dashboards, detects anomalies using LLM reasoning, and generates actionable incident reports — all without human intervention.

How It Works

1. Monitor: Agent polls streaming metrics via Grafana MCP tools (check_stream_health, detect_anomalies)
2. Detect: Gemini 3.5 Flash analyzes patterns across bitrate, latency, error rates, and CDN health
3. Alert: Creates Grafana dashboards programmatically when anomalies are detected
4. Report: Generates structured incident reports with root-cause analysis and recommended actions

Unique Angle

Unlike traditional monitoring tools that use static thresholds, Cinema Sentinel uses Gemini's reasoning to understand the context of anomalies — distinguishing between a CDN failover (expected) and a cascading failure (critical). The MCP-based architecture means any AI agent can call these tools autonomously.

Tech Stack

• Google ADK — Agent framework with LlmAgent pattern (reason→tool→observe→respond)
• Gemini 3.5 Flash — Real-time anomaly reasoning
• Grafana MCP Server — 5 MCP tools for observability data
• Google Cloud — Firestore (mock SQLite), Pub/Sub (mock asyncio)
• FastAPI — Backend with WebSocket for real-time updates
• Vercel — Free-tier deployment

5 MCP Tools

1. check_stream_health — Query current bitrate, latency, error rates
2. detect_anomalies — Identify patterns deviating from baseline
3. create_grafana_dashboard — Auto-generate incident dashboards
4. generate_alert — Send contextual alerts to stakeholders
5. produce_incident_report — Structured root-cause analysis report

Demo Flow

Run POST /api/demo to trigger a full monitoring cycle:

• Agent polls 3 simulated stream endpoints
• Detects a bitrate drop anomaly on Stream-2
• Creates a Grafana dashboard for the incident
• Generates an incident report with severity + root cause
• Sends an alert notification

What's Real vs Mocked

• ✅ REAL: Google ADK agent framework (genuine LlmAgent), FastAPI backend, MCP tool architecture, SQLite persistence, async event loop, all REST endpoints
• ✅ REAL: Gemini integration (mock mode when no API key, real when key provided)
• 🔧 MOCKED: Grafana MCP responses (simulated metrics), Firestore (SQLite), Pub/Sub (asyncio queues)

Grafana Partner Track Fit

Cinema Sentinel is purpose-built for the Grafana track — it uses Grafana MCP as its core data source and auto-generates Grafana dashboards as its primary output. The agent's entire value proposition centers on making Grafana observability data actionable through AI reasoning.

Built With

• ai-agents
• fastapi
• gemini
• google-adk
• google-cloud
• grafana
• mcp-server
• python
• vercel