Greenlight

AI agents turn screenplays into production-ready intelligence surfacing permits, safety risks, dependencies and mitigations before they become costly production blockers.

Comment

Greenlight

Screenplay intelligence → production decisions

A screenplay tells you what happens on screen.

It doesn't tell you whether you can actually shoot it.

A drone chase over a city street. A night shoot involving pyrotechnics. A glacier sequence in Svalbard. A firearm on a public location.

Each can introduce permits, safety requirements, jurisdictional rules, lead times, specialist crew, and production blockers.

Greenlight turns the screenplay itself into a production-readiness assessment.

What inspired us

Film production teams often discover operational constraints after creative decisions have already been made.

We wanted to move that intelligence upstream.

Instead of asking:

"What could go wrong when we get to production?"

Greenlight asks:

"What does this screenplay require before we commit to production?"

What Greenlight does

Give Greenlight a screenplay—or choose one of our demo scenarios.

Our agentic pipeline:

INGEST → EXTRACT → TRIAGE → RESEARCH → ASSESS → MITIGATE → GREENLIGHT

It:

  1. Extracts scenes, locations, actions, and production elements from the screenplay.
  2. Triages those elements into production dependencies and hazards.
  3. Researches relevant external evidence for the applicable jurisdiction.
  4. Assesses each dependency using severity, evidence confidence, context, and regulatory signals.
  5. Prescribes mitigations that can turn production exposure into an actionable plan.
  6. Generates an explainable production decision—GREEN, YELLOW, or RED.

The result is more than a risk list.

It is a Greenlight Production Brief showing:

  • Overall production readiness
  • Critical blockers and top risks
  • Jurisdiction applied
  • Evidence and citations
  • Recommended mitigations
  • Scene-level production impact
  • A complete decision chain
  • An explainable score showing exactly how readiness was calculated

Why it is agentic

Greenlight is not simply an LLM reading a screenplay and producing a paragraph.

The system coordinates multiple reasoning stages and external research:

Screenplay → Production Dependencies → Jurisdiction → Evidence → Assessment → Mitigation → Decision

The research stage looks for jurisdiction-qualified evidence, while the scoring engine remains deterministic and explainable.

That separation matters.

The AI helps discover and interpret production intelligence.
The deterministic engine makes the final readiness calculation reproducible.

The scoring model

Every dependency contributes to cumulative production exposure.

The system starts at:

100 — READY

Risk contributions reduce readiness based on factors such as severity, evidence confidence, scene context, jurisdiction match, and lead time.

Certain regulatory findings can also create a hard-stop ceiling, preventing a high-risk production from appearing deceptively ready simply because other areas are low risk.

Every deduction can be reconstructed from the audit trail.

For example:

100 → 85

or:

100 → 40 → 26 → 19 → 14 → 11

The judge doesn't have to trust a black-box number.

Greenlight shows how it got there.

What we built

Greenlight includes:

  • A screenplay parser and production-element extractor
  • A catalogue of production hazard classes and contextual rules
  • Jurisdiction and authority matching
  • External evidence research and source binding
  • Evidence confidence and provenance tracking
  • A deterministic explainable scoring engine
  • Mitigation generation
  • An agent orchestration layer
  • A real-time streaming activity interface
  • An explainability and audit-trail UI
  • Curated scenarios and regression tests

The system includes 193 automated tests covering scoring, extraction, evidence binding, agent workflow, API/SSE behavior, frontend explainability, and regression scenarios.

What we learned

The hardest part wasn't getting an AI model to identify hazards.

It was making the result trustworthy enough to support a production decision.

That led us to separate probabilistic intelligence from deterministic decision logic, bind findings to evidence, expose provenance, account for jurisdiction, and make every score explainable.

The vision

Greenlight is a prototype for a broader idea:

Creative intelligence should connect directly to operational intelligence.

Before a production reaches the location manager, safety team, legal team, or production office, Greenlight can surface the questions that need answers.

Because the best production blocker is the one you discover before production.

Built With

Share this project:

Updates

Submission history