Real-Time Video Agents Hack
Explore the projects and technologies built at the hackathon.
Showing 1–30 of 45 projects
seth
Aloha Circle turns arriving on Maui into meeting Maui — travelers scan a QR at baggage claim, share a playful aloha ritual with an AR guide, and get matched with a real local to enjoy a donated island experience together, all while supporting Maui causes.
james.bian
Yield or proceed use Cosmos reason to label warehouse footage indexed on Vast to train a tiny CPU model
Zohreh Dannenhauer
Property Studio turns real estate listings into attention-grabbing video scripts backed by photo analysis, property research, and transparent financial models.
Rowan Cooper
This is a project that uses real time video to detect obstacles on sideawalks for disabled people. Generally, systems that do this track permanent obstacles, but this is for temporary ones.
Bindu Bhargava Reddy Chintam
ViZ Agent finds pedestrian–vehicle conflicts in street and dashcam video, double-checks each one with three AI checks, and turns the verified ones into ranked hotspots and a drafted work order for city traffic engineers.
Rajat Mahesh Gupta
Praxis — Find my Car is a video agent for people who lose a car in dense street parking or residential blocks. Instead of scrubbing hours of camera footage, you ask in natural language.
priyanshudangipd
OriginX lets site safety leads write safety rules in plain English and watches the video archive for them, raising scored incidents with written reports.
André Antonelli
Evidence Atlas puts GPS-less dashcam footage on the map using only what the camera saw. Cosmos Reason reads the signs, OpenStreetMap supplies the coordinates, and a route-continuity check rejects impossible matches. Every pin shows its evidence.
Yuchia Chang
We changed our project to a crosswalk block detector that speak out the warning to driver who stopped on crosswalk.
Jacob
Describe a dangerous moment in one sentence; Grabbit digs through pre-indexed warehouse and road footage and hands you a verified, exportable test set of edge cases for robotics/AV policy evaluation. AI proposes, a human decides.
Christopher Donaldson
Near-Miss Board lets traffic safety teams search hours of multi-camera highway footage in plain English to find vehicles in near-miss close calls.
Karanjit Singh
turns a plain language request into a clean versioned training dataset
ankur.a.agarwal
Warehouse Safety Monitor watches warehouse camera footage and gives shift leads a ranked queue of near misses, blocked aisles, crowding and unsafe behavior, with measured distances, hotspot heatmaps and an end-of-shift report.
georgeshostakovich
StreetStyle Scout searches street footage for visible clothing and colors, then shows matching frames and video clips so you can verify the results.
Anonymous builder
Perjury is a fact-checker for traffic video: you say or type a claim about the footage, and it splits the sentence into atomic claims, checks each one, and rules each SUPPORTED, CONTRADICTED, or "the pixels can't tell" with the evidence shown.
Gridlock Box LLC
Integrating Andon with Vision Training to Develop Safe Workflows
Anonymous builder
VAST CAM uses VAST AI OS to build targeted video datasets for training camera-specific models that predict warehouse collisions before impact.
Xinshuo Jiang
A reviewer searches indexed footage in plain language and gets ranked evidence cards linked to the source interval.
Manvender Raparia
WORLDSTATE learns how physical processes should unfold from repeated video, detects when reality deviates, explains failures with NVIDIA Cosmos, and remembers them so future incidents are recognized instantly. Built for robotics, manufacturing, and warehouse operations.
Anonymous builder
MomentOS turns dashcam history into an explainable driving risk score, using video evidence to identify near-misses, attribute cause, and help insurers reward safer drivers.
Touko Ursin
Understudy turns video nobody watches into a live coach. It learns a task from expert recordings, then watches a new person on camera, corrects mistakes out loud with the expert clip for that step, and scores the attempt.
Anonymous builder
Eyes on the Floor is a video agent that watches a warehouse's existing cameras, flags OSHA hazards like blocked exits and people in forklift lanes the moment they appear, and alerts the floor lead in seconds, so hazards get fixed before an injury, a complaint or an inspection.
Anonymous builder
CityPulse lets you ask a city a natural-language question and receive a grounded 30-second narrated video answer assembled from the most relevant moments in its indexed footage.
Alexander Skinderev
EqualEd is a classroom video agent for neurodivergent students: it sees sensory overload coming for an autistic student and starts calming sound before the noise, rewinds what a student with ADHD missed, and alerts the teacher only when they need to step in.
vaishnav.ajai.th
A video agent for rideshare drivers: coaching from their dashcam clips, evidence for unfair ratings.
Anup Ghatage
Video aided gaussian splat style 3D reconstruction of any incident from any video for insurance claims
Kaushik Sivakumar
Hop24 is a game where you guide a chicken across real highway footage, avoiding vehicles detected by AI.
kangu10
Give your legacy app a second life — add an agent, not a rewrite. It learns an ERP workflow from one screen recording, then creates real invoices in the live app using only what it can see on screen, and hands the keyboard back whenever it is unsure.
Anonymous builder
LongTail finds the rare, risky moments hidden in hours of unwatched traffic and warehouse video, double-checks each with an AI, and shows what the archive can't see, for self-driving and robotics teams that need edge cases to test on.
Gabriele Battimelli
Video Scrubbing Tool