Foresight

AI Rescue Platform to Locate Missing Children in Minutes, Not Hours.

Comment 1

📌 Inspiration

According to the U.S. Department of Homeland Security, a child goes missing every 40 seconds in the United States.

Despite the magnitude of this crisis, the vast majority of cases remain unsolved—mainly due to a lack of fast, actionable information for first responders. Without timely intel, cases can take weeks, months, or even years to resolve.

🔍 What It Does

Foresight is an AI-powered platform that radically transforms how missing persons—especially children—are located.

  • Processes live CCTV footage in real-time (under 2 seconds latency)
  • Uses YOLOv11 to detect humans and extract features like clothing, age, accessories
  • Integrates with Google Gemini to generate natural-language descriptions
    (e.g., “teenager in blue hoodie holding a backpack”)
  • Displays alert matches against scraped Amber Alert databases

▶️ Live Demo

Watch the pipeline in action

🛠️ How We Built It

🧠 Backend

  • FastAPI + Uvicorn — Lightweight, real-time API server
  • YOLOv11 — Human and object detection from video frames
  • Google Gemini — Attribute recognition & description generation
  • OpenCV — Frame parsing and filtering
  • MongoDB Atlas — Cloud-based video metadata and vector storage

💻 Frontend

  • Next.js — Blazing fast web interface
  • TailwindCSS — Utility-first styling
  • ShadCN — Modern UI component library
  • Leaflet — Interactive camera map
  • Framer Motion — Smooth animations and transitions

🧩 Interface Features

🧠 RAG Chat + Smart Search

Foresight integrates a Retrieval-Augmented Generation (RAG) model for chatting with the system and narrowing search parameters intuitively.

RAG sidebar chat UI

🟥 Amber Alert Integration

Foresight scrapes real-time Amber Alert databases and lets users query for missing children by physical traits or metadata, instantly surfacing related cases.

Amber Alert Dashboard

⚙️ Challenges We Faced

  • Reducing pipeline latency to keep detection under 2 seconds
  • Handling multi-stream video ingestion on the front end
  • Aligning object detection output with semantic search relevance

✅ Accomplishments We're Proud Of

  • Matched live video frames to user prompts and photos within the first 5 hours
  • Designed an intuitive, clean interface for high-pressure search operations
  • Fully optimized AI search pipeline for fast, meaningful visual results

📚 What We Learned

  • How to optimize Gemini to extract meaningful visual details from live video
  • How to use vector search for probabilistic matching in real-world frames
  • Best practices for building a real-time visual intelligence dashboard

🚀 What’s Next for Foresight

  • Expand camera network access across California through municipal partnerships
  • Integrate real-time 911 call transcription and sentiment detection

- Further optimize semantic visual search for even faster matching

Thanks for reading! <3

Built With

Share this project:

Updates