RightScan

Autonomous AI multi-agent intelligence for trademark and rights clearance across production video, audio, imagery & scripts.

Comment

Inspiration

I've always been drawn to how brands, trademarks, and copyright actually work behind the scenes - the legal machinery most people never think about until it becomes a problem. Rights clearance is one of the least glamorous parts of film production, but one of the most expensive: every visible logo, mentioned brand, or recognizable trademark in a finished film has to be manually reviewed before release, or the studio risks real legal exposure. This process is still largely manual, slow, and easy to miss things in - especially for smaller productions without a dedicated legal team.

When I saw this hackathon's challenge, it felt like a natural fit: an AI agent that could scan a script alongside its video and audio, and give production teams a fast, structured first pass on clearance risk - catching not just what's visible on screen, but what's said out loud, before it ever reaches a human reviewer. The goal isn't to replace legal judgment, but to give production crews a faster, more confident starting point.

What it does

RightScan is an automated multimedia rights clearance platform powered by autonomous AI agents and high-performance analytics:

  • Multimodal Asset Ingestion: Upload scripts, production stills, audio stems, and high-resolution video footage.
  • Autonomous Clearance Agent: An AI agent orchestrates multi-tool pipelines to detect visual logos, parse dialogue for copyrighted entities, identify songs, and evaluate brand prominence.
  • Exact Timestamp & Frame Extraction: For videos, RightScan pins down the exact second a third-party brand appears, dynamically capturing preview frames and rendering high-contrast bounding boxes around the marks.
  • Legal Clearance Rationales & Risk Scoring: Categorizes each exposure into High, Medium, or Low risk based on screen duration, narrative role, and commercial prominence—providing entertainment-law-backed rationales for fair use, trademark dilution, or required licensing.
  • Audit History & Analytics: Stores complete multi-run clearance audit trails, detection metrics, and time-stamped logs for insurance underwriting and chain-of-title documentation.

How I built it

  • Multimodal AI Reasoning: Built with Google Gemini 2.5 Flash and the Google Agent Development Kit (ADK) to perform visual entity extraction, frame-level logo identification, bounding box coordinate mapping, and audio dialogue transcription.
  • Media Processing & Dynamic Snippets: Integrated Cloudinary for video and image storage, utilizing server-side start-offset thumbnail transforms (so_timestamp) to dynamically generate video frame snippets at the exact second a brand appears.
  • High-Speed Clearance Database: Powered by ClickHouse Cloud for analytics, storing project assets, detection coordinates, risk breakdowns, agent tool calls, and immutable audit logs.
  • Architecture & Tooling: Built with a Node.js/Express TypeScript backend leveraging Model Context Protocol (MCP) tool patterns (extract_script_entities, detect_visual_logos, transcribe_and_flag_dialogue, score_risk) and a React + Vite frontend styled with Tailwind CSS, featuring retro-styled high-contrast legal risk cards.
  • Production Deployment: Containerized with Docker for full-stack deployment on Render.

Challenges I ran into

  • Multimodal Time Synchronization: Mapping brand detections in video files to precise visual frames without forcing the browser to download gigabytes of raw video. We solved this by pairing Gemini’s temporal cues with Cloudinary's dynamic on-the-fly frame extraction API.
  • Coordinate Mapping & Bounding Boxes: Normalizing coordinate spaces between Gemini's box_2d outputs (0–1000 scale) and diverse aspect ratios (16:9, anamorphic, vertical formats) so bounding boxes accurately highlight brand marks across all viewports.
  • Agent Deduplication & Scalability: Ensuring that multi-asset projects with dozens of overlapping brand mentions across scripts and video frames were deduplicated per asset while preserving distinct timestamps and frame evidence.

Accomplishments that I am proud of

  • Zero-Download Frame Previews: Delivering instant visual proof with exact time codes and bounding boxes so legal teams can inspect violations in seconds without scrubbing through long video timelines.
  • Comprehensive Multi-Modal Coverage: Successfully handling three distinct media types—visual footage, spoken audio tracks, and written screenplay documents—under a unified legal clearance dashboard.
  • Production-Grade Architecture: Seamlessly uniting ClickHouse Cloud, Cloudinary, and Google GenAI into an agentic workflow that feels instantaneous and reliable.

What I learned

  • The Nuance of Legal Clearance in AI: Evaluating brand risk isn't just binary object detection—context matters tremendously. A logo featured centrally on a main character has vastly different legal exposure than an incidental brand blurred in background traffic. Guiding our agent prompts to evaluate narrative role, camera prominence, and trademark law factors made the outputs genuinely practical.
  • Agentic Orchestration with ClickHouse: Streaming structured agent tool calls and detection metadata into columnar tables makes querying clearance statistics and running diffs between project cuts blazing fast.

What's next for RightScan

  • Automated Blur & Inpainting Workflows: Adding one-click AI video inpainting to digitally remove or replace flagged logos directly from the report.
  • Direct Cloud Storage Connectors: Native Google Drive, Frame.io, and Dropbox integrations for automatic ingest straight from production cameras and editing suites.
  • Music Licensing Database Integration: Live cross-referencing with ASCAP, BMI, and SESAC databases to verify master recording rights and publishing splits.
  • Studio Export Packages: One-click generation of comprehensive PDF/E&O insurance clearance binders formatted for studio distribution and film festivals.

Built With

Share this project:

Updates

Submission history