CinePulse OS

💡 Inspiration

In the modern entertainment landscape, film studios and streaming platforms invest hundreds of millions of dollars producing blockbusters and episodic series. However, they face a critical blindspot: by the time traditional analytics teams compile audience drop-off reports weeks after a premiere, millions in viewer retention and box office momentum have already been lost.

We were inspired to build CinePulse OS—an autonomous Hollywood Studio Control Room where filmmakers, showrunners, and studio executives can command a multi-agent AI swarm in plain English to diagnose streaming churn, analyze scene-by-scene telemetry, and simulate theatrical vs SVOD financial elasticity in under 10 milliseconds.

🎬 What It Does

CinePulse OS is an enterprise-grade agentic intelligence platform integrating Google Gemini 2.5 with a Sub-10ms ClickHouse Cloud database across 4 core modules:

1. 🤖 Multi-Agent Studio Swarm:
   • 🎬 The Director Agent (Gemini 2.5 Flash): Deconstructs complex creative and business queries into data retrieval hypotheses.
   • 📊 The Technical Producer (ClickHouse SQL Engine): Scans millions of streaming telemetry events, buffering logs, and box office records in sub-10ms.
   • 🏛️ The Studio Head Agent: Synthesizes raw analytical findings into high-ROI Hollywood decisions (e.g., scene re-edits, marketing budget shifts, CDN edge caching).
2. ✂️ Interactive "Director's Cut" Retention Simulator:
   • An interactive slider suite allowing directors to simulate trimming exposition dialogue and boosting scene pacing, dynamically recalculating the before-and-after retention curve and annual subscriber value saved ($M USD).
3. 🎞️ 4 Proprietary Cinematic Data Engines:
   • Frame-Level Script-to-Telemetry Alignment: Maps viewer drop-off timestamps directly to script shot types (Close-Up, Wide Shot) and narrative themes.
   • Screen-Space Visual Cognition Heatmapper: Detects how low screen luminance (<45 Nits) on mobile OLED displays causes viewer abandonment.
   • Real-Time Perceptual Pacing Hash Pipeline: Correlates rapid cut frequency (cuts/min) and audio dB variance with CDN buffer stress.
   • Theatrical vs SVOD Elasticity Matrix: Simulates direct-to-streaming subscriber acquisition value vs theatrical windowing ROI.
4. 📄 1-Click Executive Studio Memo Export:
   • Exports formatted, boardroom-ready executive intelligence briefs for investor meetings.

⚙️ How We Built It

• Google Cloud AI: Built with the new google-genai SDK using Google Gemini 2.5 Flash for multi-agent reasoning, intent extraction, and NL-to-SQL generation.
• ClickHouse Cloud: Hosted high-performance analytical cluster in AWS Mumbai (ap-south-1) storing 8 cinematic telemetry and box office tables via clickhouse-connect, backed by an automatic in-memory fallback engine.
• Enterprise UI/UX: Crafted with a pitch-black cyber-cinema aesthetic (#000000), obsidian glassmorphism cards, dynamic SVG sparklines, and Plotly data visualizations.
• Backend Architecture: Production FastAPI microservice with REST endpoints for swarm orchestration and simulation.

🧗 Challenges We Ran Into

• Sub-10ms Telemetry Query Optimization: Streaming telemetry tables generate millions of timestamped drop-off logs. We designed optimized ClickHouse MergeTree engines sorted by (title, session_id) to ensure multi-agent queries complete in single-digit milliseconds.
• Bridging Creative & Technical Contexts: Data engineers care about bitrates and milliseconds, while directors care about character arcs and emotional pacing. Prompting Gemini with structured Hollywood director personas bridged this gap seamlessly.

🏆 Accomplishments That We're Proud Of

• Achieving sub-10ms analytical query speeds on live ClickHouse Cloud.
• Developing the Director's Cut Simulator that makes complex statistical retention modeling intuitive with interactive sliders.
• Delivering a world-class, AAA-tier dark command terminal UI that looks and feels like a real Hollywood production control room.

📚 What We Learned

• How combining Google Gemini 2.5 with ClickHouse's columnar speed enables true real-time agentic workflows that would be impossible with traditional relational databases.
• The power of multimodal and perceptual metrics (screen luminance, shot types) in predicting audience engagement before a movie ever hits theaters.

🚀 What's Next for CinePulse OS

• Gemini 2.5 Multimodal Video Ingestion: Direct frame-by-frame raw video rendering analysis.
• NLE Plugin Integrations: Native plugins for Adobe Premiere Pro and DaVinci Resolve to export automated AI rough cuts directly onto the editor's timeline.

Built With

• clickhouse
• clickhouse-cloud
• fastapi
• gemini
• google-cloud
• pandas
• plotly
• python
• streamlit