Aegis - Project Details

Back to Projects

Aegis

Hackathon · PineQuest Ep.2

AI Property-Monitoring Platform

View on GitHub
2 Weeks
5 Developers · Team Lead
Applied AI / Computer Vision
Up to 30 concurrent camerasHigh-confidence filtering (≥0.7)
Project Overview

Aegis is an AI-powered monitoring platform that automatically detects residential common-area violations such as smoking, vaping, and littering inside apartment complexes. It pairs low-latency edge detection (YOLOv8) with deep vision reasoning (Google Gemini 2.5 Flash) to catch violations accurately while keeping cloud costs low. Built for apartment homeowners' associations, property managers, and building security teams.

Key Features
Two-stage AI vision pipeline (YOLOv8 edge gate → Gemini verification)
Automatic LAN camera discovery (ONVIF + TCP port scans)
Multi-stream RTSP ingestion, up to 30 concurrent cameras
Fault-tolerant streaming (auto-reconnect, frame-skip, 60s backoff)
High-confidence event filtering to Cloudflare D1 / R2
Real-time violation dashboard
Tech Stack
YOLOv8Google Gemini 2.5 FlashLitServeNext.jsTypeScriptCloudflare D1Cloudflare R2FFmpegRTSP / ONVIF
Challenges & Solutions

Challenge

High cloud-API cost from analyzing every frame

Solution

A lightweight YOLOv8 edge gate screens frames first; only high-confidence crops are forwarded to Gemini for deep analysis.

Challenge

Unreliable camera networks and dropouts

Solution

A fault-tolerant layer with automatic reconnection and a 60-second backoff keeps the dashboard live without crashing.

Challenge

Discovering cameras with zero configuration

Solution

ONVIF plus custom TCP port scanning auto-detects active IP / NVR cameras for a plug-and-play setup.

Team Roles

Danny

Team Lead — architected the two-stage AI vision pipeline, camera discovery, and RTSP ingestion; led a team of 5.