Sentinel

Our project uses AI/ML to monitor public spaces in real-time, frame-by-frame, detecting instances of violence(ex:crowds)to prevent harm and promote accountability. We aim to foster safer communities.

Comment

Inspiration

Sentinel was born from a simple question: What if AI could witness injustice in real time?
In an age of rising concerns around police brutality, protest violence, and accountability, we wanted to create a tool that watches with purpose. Inspired by movements for justice and the power of transparency, we aimed to build an AI that doesn't just analyze footage — it protects our communities.

What it does

Sentinel is a real-time AI system for civic safety. Users upload videos of public interactions (like protests or police stops), and Sentinel:

  • Flags potentially violent or aggressive behavior frame by frame
  • Captures and ranks flagged frames by severity (based on violence confidence level)
  • Uses Gemini to generate a summary of relevant civil rights
  • Displays flagged frames with timestamp, GPS (if available), and contextual metadata

How we built it

  • Frontend: HTML, CSS, JavaScript — designed with responsiveness and accessibility in mind
  • Backend: Flask API for video handling, frame extraction, and model inference
  • Model: Custom-trained violence detection model using Kaggle video data, adapted for frame-level prediction
  • AI Summary: Gemini API generates civil rights summaries based on interaction context
  • Computer Vision: OpenCV for frame-by-frame video processing and metadata tagging

Challenges we ran into

  • Debugging model input shape issues when converting video into image sequences
  • Finding a realistic, diverse dataset to train on real-world violence scenarios
  • Tackling Flask bugs, TensorFlow errors, and circular imports
  • Integrating Gemini AI summaries in a user-friendly interface
  • Managing live feedback and notifications without breaking the UI
  • Git version control conflicts during collaborative development

Accomplishments that we're proud of

  • Created a fully functional real-time violence detection pipeline with actionable rights feedback
  • Built a custom-trained violence classifier tailored to civilian-police encounters
  • Designed a professional UI that makes legal education intuitive and accessible
  • Seamlessly integrated Gemini summaries to translate legal jargon into plain English
  • Demonstrated that AI can be used not just for surveillance, but for empowerment and justice

What we learned

  • How to turn an AI concept into a full-stack, deployable application
  • The importance of custom datasets vs. off-the-shelf models
  • Why modular code and clean file structure are essential — especially with Flask
  • How impactful AI can be when pointed at real-world problems

What's next for Sentinel

  • Integrate with city surveillance feeds for real-time alerting
  • Add multilingual rights summary support (especially for migrant communities)
  • Expand detection to include verbal intimidation and threats via audio
  • Partner with civil rights organizations to bring Sentinel to communities that need it most
Share this project:

Updates