Inspiration
Scripts often include real-world facts, people, events, locations, brands, and news references. Even a small mistake or unverified reference can create credibility, legal, or rights-related problems. I built ScriptShield to help creators identify these risks before production or release.
What it does
ScriptShield analyses an uploaded script and identifies doubtful statements or portions that require verification. It checks factual and news-related claims, highlights potential legal and rights concerns, and classifies each finding as: Safe/ Needs Clearance/ Factual Error/ Legal Risk.
The final report includes explanations, evidence, and recommended actions.
How I built it
I built ScriptShield as a sequential agent workflow using Gemini and Parallel Search:
Extractor Agent: Extracts doubtful claims and risky script portions.
Verifier Agent: Uses Parallel to search for reliable sources and collect supporting or conflicting evidence.
Classifier Agent: Evaluates the evidence and assigns an appropriate risk category.
The system then presents the results in a structured verification report for human review.
Challenges I ran into
The biggest challenge was handling ambiguous, incomplete, outdated, or jurisdiction-dependent information. Some claims could not be confidently classified as true or false. To address this, I added confidence levels and a Needs Clearance category instead of forcing uncertain findings into a definitive answer.
Accomplishments that I'm proud of
I created a working agentic workflow that connects script analysis, web-based verification, and risk classification into one practical tool. I'm especially proud of making the results evidence-backed and actionable rather than simply generating unsupported opinions.
What I learned
I learnt how to design sequential agents, pass structured outputs between stages, use external search for verification, and distinguish factual accuracy from legal-risk detection. I also learned that AI should flag potential legal issues and not provide a final legal approval.
What's next for ScriptShield
As future steps, I plan to add PDF and screenplay-format support, source credibility scoring, citation export, multilingual scripts, and integrations with production workflows. I also want to include human feedback in loop so editors and legal teams can improve the system over time.
Log in or sign up for Devpost to join the conversation.