StudioClear: Evidence-backed script clearance before production
Inspiration
Film and television scripts frequently reference real people, brands, organizations, locations, products, and events. Each reference can create clearance questions, but researching them manually requires production teams to search across disconnected sources, preserve evidence, evaluate uncertainty, and coordinate legal review.
This work is often performed late in production, when changes are expensive.
We built StudioClear to move clearance research earlier in the creative process. It gives writers and production teams a fast, transparent way to identify potentially sensitive references, investigate them using current sources, and determine which items require human review.
What it does
StudioClear transforms an uploaded screenplay into an evidence-backed clearance report.
The workflow:
- Gemini extracts references from the screenplay while preserving the exact source text.
- Google ADK agents plan and coordinate the research workflow.
- Parallel Search is called at runtime to investigate each reference using current web sources.
- Every finding is connected to its citations, evidence, and reasoning.
- A deterministic policy engine classifies each item as:
- CLEAR — available evidence does not trigger a clearance rule.
- REVIEW — uncertainty or policy conditions require human review.
- ESCALATE — the reference presents a significant unresolved concern.
- An authorized reviewer can override a recommendation, and StudioClear records the decision in an audit trail.
StudioClear does not pretend to replace legal judgment. It prepares a traceable research packet so production and legal teams can make faster, better-informed decisions.
How we built it
StudioClear is a multi-agent application deployed on Google Cloud Run.
- Gemini performs structured screenplay analysis and reference extraction.
- Google Agent Development Kit (ADK) orchestrates specialized planning, research, and review agents.
- Parallel Search supplies live, cited web research during every analysis run.
- A deterministic Python policy engine converts the collected evidence into consistent clearance recommendations.
- The web interface supports screenplay upload, evidence inspection, clearance reporting, human override, and audit verification.
The architecture deliberately separates responsibilities:
- AI agents extract references and gather evidence.
- Parallel provides current research with source traceability.
- Policy code produces consistent recommendations.
- Human reviewers retain authority over the final decision.
The application was validated with 45 automated tests, including workflow, policy, authorization, and audit behavior.
Challenges we faced
The greatest challenge was turning search and model output into something a production team could trust.
A list of search results is not a clearance decision, while a simple LLM-generated “safe” or “unsafe” label is difficult to verify. We therefore had to preserve the connection between the screenplay text, research query, evidence, citation, policy rule, and final recommendation.
Other challenges included:
- distinguishing verified evidence from incomplete or conflicting information;
- preventing unsupported model conclusions;
- coordinating multiple agents without losing traceability;
- keeping deterministic policy enforcement separate from probabilistic reasoning;
- supporting human overrides without erasing the original recommendation; and
- presenting a complex research workflow through a simple product experience.
What we learned
We learned that trustworthy agentic systems require more than capable models. They need clear responsibility boundaries, evidence provenance, deterministic controls, and human accountability.
Gemini and Google ADK made it possible to understand and orchestrate the screenplay workflow. Parallel made the output useful by providing current, source-backed research. The deterministic policy layer converted that evidence into consistent and explainable recommendations.
Most importantly, we learned that the strongest agentic workflow does not hide uncertainty. It shows users what the system found, what remains unresolved, why a decision was recommended, and when a human must take control.
What's next
Next, we plan to add:
- studio-specific clearance policies;
- collaborative review queues for production and legal teams;
- screenplay version comparison and change tracking;
- rights-holder and licensing workflows;
- configurable jurisdiction and production policies; and
- integrations with studio production-management systems.
Our goal is to make clearance a continuous, evidence-backed part of the creative workflow—not a last-minute production bottleneck.
Log in or sign up for Devpost to join the conversation.