Submissions from builders giving agents video, audio, memory, and actions with VideoDB.
On June 13, 2026, 100+ builders gathered at SMU School of Economics in Singapore for Agent Forge, a one-day in-person hackathon hosted by AI Builders and SMU AI. VideoDB partnered alongside Kimi AI, Bright Data, and Daytona to bring the perception stack.
The format was simple. Teams had from morning to evening to build a working AI application. Our lens was specific: use VideoDB to give agents the ability to see, understand, and act on video and audio. By sunset, 50 projects had shipped. We picked 8 that pushed VideoDB the furthest.
Tracks Arsenal FC players at World Cup 2026 with semantic goal search, player profiles, and HLS supercuts of match highlights.
Uses spoken-word indexing and semantic search to compile exact-match highlight reels from match footage.
An autonomous insurance pipeline that ingests dashcam footage, detects collisions, identifies fault, and produces an approve-or-reject decision with evidence clips.
Uses structured scene indexing prompts and spoken-word indexing for auditable, clip-backed claim decisions.
Turns live-selling streams into shoppable vertical clips with auto-timed cuts, caption overlays, and a ready-to-post product gallery.
Uses spoken-word search, the Timeline API for programmatic editing, and CaptionAsset for auto-captions.
A competitive intelligence platform that ingests brand videos, performs dual-layer search over transcripts and scenes, and surfaces cross-brand analysis reports.
Uses transcription, spoken-word and scene indexing, dual-layer semantic search, and HLS streaming.
A Telegram bot that routes citizen photo and video reports to the correct government authority with AI-identified problem type and severity.
Uses VideoDB for evidence storage, async polling, and scene indexing with civic-detection prompts.
A human-in-the-loop CCTV dashboard that detects fights via scene indexing and returns streamable evidence clips for operator review.
Uses time-batched scene indexing with fight-detection prompts and webhook alert integration.
A multi-agent teacher evaluation pipeline that analyzes lecture recordings, evaluates teaching quality, and cross-references gaps with student exam data.
Uses transcription with sentence segments, spoken-word indexing, semantic search, and clip generation across a LangGraph pipeline.
A mobile food discovery app that indexes trending creator clips for food-specific search and builds a swipeable dinner shortlist from evidence-backed picks.
Uses dual indexing with food-specific prompts and semantic search across audio and visual index types.
