Greenlight AI

Inspiration

The core inspiration behind this project was to push my technical boundaries by architecting a multi-agent AI system. I wanted to design a collaborative network of AI Agents that functions as a cohesive team to help creators brainstorm, refine, and validate new film ideas.
To solve the common industry problem of AI hallucinations, the multi-agent system is mathematically grounded in real-world data constraints:
$$ V(Idea) = A_{orchestration}(f(MarketData, HistoricalTrends)) $$

This ensures that every generated concept is deeply anchored in actual market realities, filtering out unrealistic or unfeasible tropes from the very beginning.

What it does

• Multi-Agent Orchestration: A core engine that delegates tasks and coordinates workflows between analytical and creative AI agents.
  
• Real-World Data Grounding: Semantic filtering that prevents hallucinated or impossible script ideas by anchoring them to historical industry data.
  
• Film Concept & Plot Generation: The system delivers a fully structured film idea, complete with a detailed narrative arc, character premises, and thematic direction.
  
• Data-Driven Marketing Analytics: Automated calculations of target audience demographics, projected market positioning, and financial feasibility metrics based on current industry trends:
  $$ MarketFit = \sum (GenreTrend \times AudienceDemand) $$

How we built it

During development, I mastered AI Orchestration, managing data streaming and state synchronization across multiple intelligent agents. The final deliverable is a fully functional MVP (Minimum Viable Product) deployed as an interactive website. The system processes user inputs, applies real-world filters, and synthesises viable cinematic concepts.

Challenges we ran into

The biggest constraint of this project was severe time limitations. Initially, the planned architecture was highly complex, which threatened the delivery of a working product.
To overcome this, the entire system architecture was completely refactored and streamlined mid-project. I pivoted from a sprawling multi-layered network to a simplified, highly efficient framework specifically designed to guarantee a working MVP.
Key Takeaways from the Pivot.

• Agile Simplification: Stripping away non-essential layers allowed me to focus heavily on the core value: grounding the AI agents in real-world data.
  
• Modular Foundation: Although simplified, the system was built with clean, modular interfaces.
  
• Ready for Scale: The current MVP acts as a robust, production-ready foundation. It is perfectly positioned for future integration of new options, advanced tools (like IBM Bob), and real-time validation features.
  

Accomplishments that we're proud of

1. High-Value Business Concept: Engineered an innovative incubation system designed to solve a tangible entertainment industry problem. By filtering out unfeasible ideas early, the platform acts as a commercial gatekeeper, saving studios immense pre-production costs and validating high-potential narratives.
   

2. Bridging Data Analytics & Software Engineering: Successfully harmonized heavy data analytics with production-ready software development. The system doesn't just run creative scripts; it seamlessly bridges data ingestion via ClickHouse with dynamic user-driven workflows, creating a unified full-stack pipeline:
   $$ SystemSynergy = DataAnalytics \times AIOrchestration $$

3. Empirical Concept Validation: Designed a robust architecture for Human-in-the-Loop validation. Even within a simulated environment, establishing a pathway where real human focus groups stream telemetry data provides an undeniable, bulletproof verification metric for movie concept viability.

4. Architectural Resilience & Adaptability: Successfully navigated high-pressure time constraints by entirely re-engineering a complex multi-layered system into a lean, highly efficient framework mid-development.

What we learned

• Delivering a fully functioning, interactive web MVP under these conditions proved my ability to prioritize value, maintain high code agility, and ship stable products.
  
• Ruthless Prioritisation: High-pressure time constraints taught me how to entirely re-engineer a complex, multi-layered system into a lean, highly efficient framework mid-development. Delivering a fully functioning, interactive web MVP under these conditions proved the value of stripping away non-essentials to ship a stable product.
  
• Designing for Future Scale: A simplified architecture does not mean a weak one. I learned to build clean, modular interfaces so that the current foundation is perfectly positioned to adopt new options, advanced tools, and automated pipelines later.
  
• AI Orchestration Dynamics: Managing multiple AI Agents taught me how to structure precise semantic constraints. I learned how to effectively ground models in real-world parameters to eliminate hallucinations and keep creative outputs deterministic.
  
• Exploring Corporate AI Frameworks: Researching IBM Bob opened up new ideas on how enterprise-grade automation tools handle complex workflows, giving me a solid blueprint for the next developmental phase.
  

What's next for Greenlight AI

Now that the core MVP foundation is stable and ready for scale, the roadmap focuses on turning this tool into an enterprise-grade validation pipeline.
The immediate backlog includes:

1. Human-in-the-Loop Focus Groups: Integrating real human focus groups to review the generated film plots and marketing strategies. The next step is to build an automated streaming pipeline to collect audience reaction data as telemetry, introducing a quantitative validation metric: $$ ValidationScore = \int_{0}^{T} Engagement(t) dt $$

2. Full IBM Bob Integration: Dedicating focused development time to fully implement IBM Bob into the orchestration layer, utilizing its enterprise automation capabilities to handle complex background agent workflows.

3. Real teaser generation by AI.

4. Interactive Marketing Dashboards: Expanding the current Grafana setup to display user-facing predictive revenue models and demographic breakdowns directly on the website platform.

5. Advanced Feature Expansion: Leveraging the current modular architecture to seamlessly plug in new user options, such as customizable budget constraints, regional trend filters, and genre-blending modifiers.

Built With

• clickhouse
• fastapi
• google-adk
• google-genai
• grafana
• mcp
• pydantic
• python
• render