Inspiration
Reaching picture lock should feel like a win, but in reality, it's where the real headache starts: music rights clearance. I’ve spent enough time around creative workflows and messy data to know how quickly manual administration kills momentum. Post teams waste days digging through clunky ASCAP and BMI registries, tracking down who owns what, and hoping the writer and publisher splits actually add up. A single missed detail can stall distribution entirely. I built CueClear to kill that bottleneck. taking an edit timeline and letting an agent handle the research, ownership math, and cue sheets automatically.
How We Built It
Timeline Ingestion: FastAPI parses .edl and .xml files from Premiere or Final Cut to grab every cue and timestamp. Agentic Research (Google ADK + Parallel): The agent searches PRO databases via Parallel Search, then uses Parallel Extract to deep-read the live registry pages. Grounded Extraction (Gemini): Gemini extracts Work IDs, publishers, writers, and shares directly from the text, tying every record to a verified search ID to prevent hallucinations. Audit & Export: An audit tool verifies that writer and publisher shares cleanly total 100%, flags disputed cues, and exports broadcast-ready Excel, CISAC XML, or JSON.
Challenges & Lessons
Zero Hallucination Tolerance: In copyright law, an AI cannot guess. We tuned Gemini to strictly report only what Parallel’s live web extracts prove. if data is missing, it flags it for human review instead of filling in blanks. Messy Database Standards: Rights registries don't format data the same way. Normalizing different split scales and handling self-published artists required strict validation logic. The Sweet Spot for Agents: High-stakes tools shouldn't be black boxes. Letting the agent automate 95% of the mechanical digging while keeping a simple dashboard for human sign-off makes the workflow fast, reliable, and trustworthy.
Log in or sign up for Devpost to join the conversation.