Product Description
GuardAngel is an AI‑powered security operations platform that turns surveillance video into prioritized, actionable intelligence. Instead of scrolling through hours of footage, operators get a real‑time stream of critical incidents, context summaries, and searchable evidence. The system is designed for safety teams to make faster, higher‑confidence decisions in campus and public safety environments.
What it Does
- Detects and ranks incidents in near real‑time from video footage
- Highlights severe cases first (medical, human‑involved, violent, collisions)
- Generates contextual summaries for each video segment
- Supports natural‑language Q&A about specific moments in footage
- Provides analytics, event history, and live incident feeds
How We Built It
- Chunked Video Pipeline: We split each uploaded surveillance video into fixed‑length chunks (e.g., ~15 seconds) using FFmpeg.
- Frame Sampling: For each chunk, we extract a small set of frames (e.g., 1 fps, capped) to keep processing fast and scalable.
- Gemini Vision: Each chunk’s frames are sent to the Gemini API, which produces a context summary and incident detections with confidence and reasoning.
- Event Storage & Streaming: Events and summaries are stored in MongoDB and streamed to the frontend using Server‑Sent Events (SSE).
- Frontend Experience: Built with Next.js + Tailwind to provide a monitoring dashboard, Q&A interface, and analytics views.
Challenges We Ran Into
- Maintaining accurate timestamps across chunked video segments
- Preventing duplicates while still ranking high‑severity incidents first
- Managing AI API reliability and output parsing
- Designing a UI that surfaces critical info without overwhelming operators
Truth & Service
GuardAngel is focused on public safety and harm prevention. It is built to support responsible decision‑making, where operators remain in control of final judgments. The system prioritizes transparency (summaries + confidence + explanation) and reduces time to response for urgent cases.
Business Model
- SaaS for campuses, municipalities, and security teams
- Tiered pricing by camera count, storage, and analytics depth
- Premium tiers for advanced analytics, multi‑camera correlation, and offline inference
Tech Stack
- Backend: FastAPI, Python
- Database: MongoDB
- AI: Gemini Vision API
- Frontend: Next.js 16, React 19, TypeScript, Tailwind CSS
- Video Processing: FFmpeg
- Streaming: Server‑Sent Events
Accomplishments We’re Proud Of
- Built a full chunk‑based video intelligence pipeline within hackathon time
- Achieved live alert streaming with triage‑first ranking
- Unified Q&A, monitoring, and analytics into one seamless platform
What We Learned
- Real‑time safety systems need strong prioritization to avoid missed incidents
- AI outputs require structured prompts, validation, and fallbacks
- Operator UX must emphasize urgency, clarity, and fast action
What’s Next
- Multi‑camera support and cross‑camera incident correlation
- Role‑based access (dispatcher vs reviewer)
- Richer analytics (trends, heatmaps, response metrics)
- Optional offline inference for privacy‑sensitive deployments
Log in or sign up for Devpost to join the conversation.