Online / July 25–27, 2026
A global media intelligence hackathon where builders turned archived footage into evidence, memory, learning, and action.
From July 25 to 27, builders worked with archived footage across research, education, compliance, public speaking, legal training, personal memory, repair, and road safety. We reviewed 26 submissions and selected ten finalists.
The ten selected projects below are shown in final order.
Ephemeris turns 87 NASA archive clips into a cited research experience. It combines speech, visual evidence, OCR, and mission metadata to trace how scientific understanding changed over time and retrieve the exact moments behind every answer.
Each answer becomes an evidence reel inside a navigable solar system, with supporting moments placed around relevant celestial bodies and mission spacecraft rendered as interactive 3D models.
CutSense turns ads, music videos, and films into a searchable reference library of editing techniques. Every result opens at the exact playable moment with a practical recipe for recreating the effect.
Its seeded library supports cross-video comparison and study reels, with recipes for VideoDB Editor, Remotion, Premiere Pro, DaVinci Resolve, and CapCut.
Turns campaign requirements into structured rules, checks creator videos against them, and gives reviewers the exact supporting or conflicting moment behind every recommendation.
Combines campaign-specific speech and visual analysis with reproducible rule checks, deterministic interval measurement, exact playback, and an immutable human review trail.
Turns courtroom archives into a searchable advocacy library where objections, rulings, contradictions, and teaching points open as exact clips instead of transcript citations.
Structures 50 hours of courtroom footage into 1,394 typed moments, then connects grounded answers, contradiction search, and multi-track teaching reels to exact testimony.
Extracts three concepts from a chapter, retrieves real educational footage for each one, and builds an interactive lesson with questions, quizzes, and personalized revision reels.
Builds each lesson from exact moments in a shared footage library, then assembles clips, narration, questions, and personalized revision reels across 11 Indian languages.
Studies the words and visible delivery of a talk, answers coaching questions with measured evidence, and automatically cuts short captioned clips that demonstrate each technique.
Connects each coaching insight to exact moments in the talk, then uses VideoDB Editor to generate short captioned examples that demonstrate the technique.
Takes a field technician's machine video from upload through diagnosis, then returns repair steps and relevant tutorial videos with jump links to the useful chapter or timestamp.
Analyzes the technician's footage alongside extracted frames and machine context, then returns a structured fault assessment with relevant tutorial chapters and timestamps.
Turns authorized farm footage into timestamped observation packets that staff can review, edit, and export for veterinary handoff without presenting visual observations as a diagnosis.
Grounds each observation in an exact clip and transcript, supports review and editing before export, and achieved an 84% pass rate in its published behavior-query evaluation.
Combines dashcam footage, depth, object detections, and behavior signals to explain a trip and deliver short audio nudges at moments where safer driving guidance is useful.
Combines VideoDB trip understanding and clip generation with local object detection and depth models, then tracks usage cost alongside every driving insight.
Makes personal video searchable by what was said, what happened, and who appeared, with a privacy review queue before sensitive footage becomes available to search.
Routes each query across speech, visual events, and recognized people, then returns the exact clip after sensitive footage passes a privacy review.
