Past eventVision Possible Hackathon
Vision Possible: Agent Protocol was a week-long online hackathon for Stream's open-source Vision Agents SDK, where developers built multimodal AI agents that watch, listen to, and understand video in real time.
- Client
- Stream (Vision Agents)
- Dates
- Format
- Online
- Prize pool
- $4,000+
Vision Possible Hackathon
Feb 23 – Mar 1 | Online
$4,000+ in prizes
Client & challenge
Stream builds Vision Agents, an open-source SDK for real-time video AI agents. It combines vision models such as YOLO, Roboflow, and Moondream with LLMs such as Gemini, OpenAI, and Claude, on Stream's low-latency edge network: 500 ms to join a call and under 30 ms of audio and video latency.
A new SDK spreads only when developers build real projects with it. Stream needed builders to ship working video agents in one week and to write about what they built.
Theme & prizes
Build a multimodal AI agent that watches, listens to, and understands video in real time with the Vision Agents SDK.
- Potential impact: a meaningful problem or a valuable Vision AI use case
- Creativity and innovation: a new idea for real-time video AI
- Technical excellence: solid use of the Vision Agents SDK
- Real-time performance: responsive and low-latency
- User experience: intuitive and polished
- Best use of Vision Agents: video AI, low latency, native LLM APIs, multi-platform
Prizes worth $4,000+
- 1st place$2,000
- 2nd place$1,500
- Best blog post$500
- Top 10 social postsSwag bundles
What we did
Partnership development
DevTools Academy initiated the hackathon with Stream, an existing client, and structured the collaboration between Stream and WeMakeDevs.
Developer reach
The collaboration connected Vision Agents with the WeMakeDevs developer community and produced 5,000+ registrations and 200+ projects built on the SDK.
Results
- Registrations
- 5,000+
- Project submissions
- 200+
- Countries
- 18+
- Prize pool
- $4,000+
Winning projects
- 1st place
ETMS: AI-powered safety infrastructure
by Rakshak
A privacy-first edge system that combines vision intelligence, anomaly detection, and health monitoring to detect falls, wandering, and unsafe situations in real time, then escalates with voice confirmation and structured emergency reports.
View code for ETMS: AI-powered safety infrastructure - 2nd place
OceanWatch: real-time marine intelligence
by Karthika M
An agent that watches and listens to marine environments through live video and audio, detects environmental pressure, and turns the signals into early risk insights.
View code for OceanWatch: real-time marine intelligence - 3rd place
Watcher: real-time posture corrector
by Marvellous
A custom processor derives body angles from YOLO results, then the agent scores your posture and gives live feedback while you work.
View code for Watcher: real-time posture corrector
Blog posts & links
Interested in running a hackathon?
Tell us about your product and the developers you want to reach. We’ll come back with a format, a timeline, and a plan to fill it.