Global Hackathon: May 2026 project showcase
A 48-hour global build event exploring what agents can do when they can see, hear, and remember video and audio.
Most AI agents only work with text. They cannot see your screen, hear your voice, or remember what happened five minutes ago. We wanted to see what happens when builders give them eyes and ears.
From May 16 to 18, 2026, VideoDB ran a global online hackathon. In 48 hours, 100 builders shipped projects that gave agents real perception over video and audio. Our judges reviewed every submission and selected the ten projects below.
01 · DataCaster
One match into four products
A football VOD or live RTSP feed becomes structured event JSON, multimodal search, an Ask interface with timestamp citations, and a vertical highlight reel delivered to Telegram.
Uses RTStream, scene indexing, spoken word indexing, voice generation, timeline editing, and sandbox compute across one media pipeline.
02 · Trace
PRs that explain themselves
A coding-session recorder that produces a narrated walkthrough, a context-aware PR description, a reviewer Q&A bot, and a contribution map.
Combines capture, upload, spoken word indexing, scene indexing, search, generated narration assets, and timeline assembly.
03 · WildWatch
AI wildlife stream monitor
A live wildlife monitor with species, behavior, environment, and audio lenses that sends Telegram alerts with playable clips when events matter.
Uses parallel video and audio reasoning so cross-modal signals can escalate live events into clip-backed alerts.
04 · ToonTalk
Make any cartoon talk back
A character interaction app that indexes cartoon scenes and dialogue, finds the right character, clones voice from the episode, and answers in context.
Retrieves the exact scene context before generating a character response and on-screen banner.
05 · GameSense
Auto clips and coaching for gamers
A gameplay assistant that detects highlights and mistakes, creates narrated vertical reels, and scores sessions across mechanics, decisions, and consistency.
Builds temporal memory over sessions so later coaching notes can point to patterns across prior gameplay.
06 · Backdrop
Search your visual memory
A native iOS memory app where short clips become searchable by plain-language prompts like a scanned book or a whiteboard from last week.
Uses scene indexing and semantic search to remove tags, folders, and manual organization from visual recall.
07 · Continuum
Prompt to mini documentary
A documentary studio that finds real YouTube footage, indexes frames and words, plans chapters, retrieves clips, generates narration, and assembles the timeline.
Layers visual indexing, spoken word indexing, audio indexing, voice generation, and programmatic editing.
08 · LobbyEar
Find political mentions that matter
A monitoring agent for hearings and proceedings that plans search angles, finds relevant mentions, explains why they matter, and returns playable clips.
Searches spoken word and visual indexes together, then adapts its strategy as evidence appears.
09 · ResearchFlow
Your research, remembered
A research memory tool that indexes lectures, screen activity, and microphone context so users can recover papers, algorithms, code ideas, and exact video moments.
Maps screen capture, dual indexing, semantic search, timeline clipping, and sandbox compute into one research workflow.
10 · DataLens
Live charts from your screen
An Electron overlay that watches screen and system audio, then turns live data patterns into charts, summaries, and anomaly alerts.
Uses capture, transcription, visual indexing, and three agents for charting, summarization, and alerting.