Inspiration
Rights clearance is often handled too late and too manually. A producer may discover a song, brand, location, artwork, or other protected element only after it is already embedded in a script, rough cut, or delivery package. The result is a trail of spreadsheets, browser tabs, emails, and uncertainty—exactly when the production needs a confident answer.
We built ClearCut to move that work upstream. It gives production teams a shared workspace where source material becomes a rights-bearing asset inventory, research becomes evidence, and evidence becomes a reviewable next action.
What it does
ClearCut accepts screenplays, source documents, video, and audio. It extracts candidate rights signals, analyzes media with Gemini on Vertex AI, researches likely ownership and licensing signals through Parallel, and produces an evidence-backed clearance card with risk, confidence, reason codes, sources, and recommended next steps.
The system keeps the consequential decision with a human. A producer, coordinator, or legal reviewer can inspect the evidence, approve or escalate an item, request more research, draft a permission request, and generate a report or branded PDF. Every important action is scoped to the production workspace and recorded in the activity history.
How we built it
The web workspace is built with Next.js, React, and TypeScript. The API is a Python FastAPI service backed by PostgreSQL through SQLAlchemy and Alembic. Firebase/Google Identity Platform provides sign-in, while the API enforces workspace membership and role-based access.
Parallel powers rights-source discovery and extraction. Google ADK and Vertex Gemini power the clearance-card agent and audiovisual analysis. Cloud Run hosts the web and API services, Cloud SQL stores application state, Cloud Storage stores source documents and media, Secret Manager protects runtime credentials, and Cloud Build publishes versioned images to Artifact Registry.
The hosted workflow is:
Source material → rights-bearing asset inventory → Parallel evidence research
→ Gemini risk and recommendation → human review → report and outreach
What we learned
We learned that useful AI workflow design is as much about boundaries and recovery as it is about model output. Gemini responses need an explicit structured schema, validation, and safe handling when a model response is incomplete. Research needs source provenance, quality signals, retryable failures, and a visible fallback that preserves evidence instead of hiding a failed generation step.
We also learned that production workflows depend on details outside the model: Firebase token verification, tenant-scoped authorization, resumable browser uploads, Cloud Storage CORS, database migrations, PDF rendering, and consistent loading and error states across the UI.
Challenges we faced
The largest challenge was making a multi-stage research workflow understandable and recoverable. Parallel could return useful evidence even when the downstream model stage failed, so we separated evidence collection from clearance-card generation and introduced a human-review fallback.
Media ingestion introduced a second class of problems: large browser uploads require resumable Cloud Storage sessions, correct CORS configuration, finalization checks, and a workflow that can represent timestamped signals. We also had to make the interface communicate what is ready for action rather than presenting a long, flat page of controls.
Outcome
ClearCut now supports an authenticated end-to-end workflow: workspace creation, invitations, role-based access, project setup, script and media ingestion, asset analysis, live Parallel research, Gemini clearance cards, human decisions, permission-request drafts, activity history, and styled report/PDF export.
ClearCut is a rights-workflow assistant, not legal advice and not a declaration that an asset is legally cleared. Final clearance remains the responsibility of the production and legal teams.
Built With
- alembic
- artifact-registry
- docker
- fastapi
- firebase-authentication
- gemini
- github
- google-adk
- google-cloud
- google-cloud-build
- google-cloud-run
- google-cloud-sql
- google-secret-manager
- identity-platform
- next.js
- parallel-extract
- parallel-search
- postgresql
- python
- react
- sqlalchemy
- typescript
- vertex-ai
Log in or sign up for Devpost to join the conversation.