Vigilens

Inspiration

Workplace safety and efficiency are still largely dependent on manual supervision, which is inconsistent, reactive, and difficult to scale. We were inspired to build a system that could act as an “AI manager”—one that continuously monitors environments, enforces standards, and helps teams improve in real time without constant human oversight. While we tailored this original iteration to the commercial kitchen environment, the underlying framework can be adapted to any high-stakes workplace where safety, compliance, and efficiency are critical.

What it does

Vigilens is an AI-powered workspace intelligence system that analyzes video feeds to detect safety violations and workflow inefficiencies in real time. It uses specialized agents to:

Identify health and safety issues (e.g., improper food handling)
Detect inefficiencies like idle time or redundant movement
Provide clear, actionable feedback through a centralized dashboard

The system adapts to different environments by learning from manager-defined policies, making it flexible across industries.

How we built it

We built Vigilens using a multi-agent architecture powered by video understanding:

TwelveLabs for semantic video analysis and event detection
Custom AI agents (Health, Efficiency) to interpret events and generate structured insights
Backend pipeline to process events and output JSON-based results
Dashboard interface to visualize violations, insights, and recommendations
Training module with Google Docs integration (OAuth) to allow managers to define custom policies

This modular design allows each component to scale and improve independently.

Challenges we ran into

Interpreting real-world video reliably and handling ambiguous scenarios
Designing agents that produce explainable, not just accurate, outputs
Structuring a flexible system that can adapt to different workplace rules
Integrating multiple tools (video AI, APIs, dashboard, OAuth) into a cohesive pipeline
Managing time constraints while building both backend logic and user-facing features

Accomplishments that we're proud of

Building a working multi-agent system that analyzes real-world activity
Creating a flexible training module that allows customization via documents
Producing structured, explainable outputs instead of black-box predictions
Designing a clean, intuitive dashboard for real-time insights
Turning a complex idea into a functional prototype within a short timeframe

What we learned

How to design and coordinate multi-agent AI systems
The importance of explainability in AI for real-world applications
How to integrate video understanding into practical workflows
Working with APIs, OAuth, and real-time data pipelines
The value of clear system architecture when building complex products quickly

What's next for Vigilens

Improve detection accuracy and handle more edge cases
Expand the agent system (e.g., compliance, security, productivity agents)
Add predictive analytics to prevent issues before they occur
Enhance the training module with better customization and feedback loops
Deploy in real-world environments like restaurants, warehouses, and healthcare settings