EvaSafe: Real-Time AI Crime Detection & Smart Surveillance

When crime happens, every second counts. EvaSafe detects it in real-time, secures the evidence on blockchain, and alerts authorities before it's too late. Surveillance that fights back !

Comment

The Moment That Started Everything

Right now, as you read this - somewhere in a city, in a street, in a country you may never know - a crime is happening.

Not yesterday. Not in a news headline from last week. Right now. This second.

A woman is being followed. A family is about to lose someone they love. And the CCTV camera on that street corner? It's recording. Passively. Silently. Doing nothing.

That reality hit me personally. The tears of families who lost their loved ones, the cases that went cold because evidence was tampered, the crimes that could have been stopped if someone — or something - had just reacted in time.

That's what kept me up at night. That's what built EvaSafe.

We Didn't Come Here to Solve One Problem

Most solutions patch one hole. We refused to do that.

Because the problem isn't just slow detection. It's also tampered evidence. It's also hacked cameras. It's also investigations that drag on for months because no one can find the right footage fast enough.

So today, we're not here to solve a single problem. We're here to break the entire chain that lets crime win.

How We Built It

We built EvaSafe as a unified intelligence platform - not a collection of disconnected tools.

  • Real-Time Crime Detection - Google Cloud Video Intelligence API + MIST model analyzes live CCTV feeds and acoustic sensors simultaneously
  • Blockchain Evidence Vault - Every incident is hash-locked on-chain via Firebase + MetaMask, making tampering physically impossible
  • Natural Language Investigation - BLIP (fine-tuned with LoRA) captions every frame, and Gemini RAG lets officers query footage in plain English: "Show me the red car near the market at 9 PM"
  • Cybersecurity Layer - Continuous CVE scanning of CCTV/DVR infrastructure so the surveillance system itself cannot be turned against us
  • Smart Alert Routing - Dijkstra's algorithm activates the 6 nearest cameras the moment a crime is detected, sealing escape routes in real time

What We Learned

We learned that technology without urgency is just a demo. Every architecture decision we made, we asked one question:

"Would this have saved someone's life tonight?"

We also learned that combining AI, Blockchain, and Cybersecurity in one coherent system is brutally hardv- but that's exactly why it hasn't been done before.

The Challenges We Faced

  • Synchronizing real-time video streams with acoustic anomaly detection without latency
  • Fine-tuning BLIP on surveillance-specific data with limited compute
  • Making blockchain evidence storage fast enough to be practically useful in emergencies
  • Keeping the entire stack production-ready under hackathon time pressure

This Is Not a Hackathon Project

This is a system that should exist in every city.

Because right now, as you finish reading this that crime is still happening somewhere.

EvaSafe exists so next time, it doesn't go unanswered.

Built With

  • acoustic
  • anamoly-detection
  • automated-cve-scanning
  • blip-(vision-language-model)
  • blockchain
  • cloud
  • computer-vision
  • cybersecurity
  • docker
  • dvr-vulnerability-detection
  • express.js
  • firebase
  • firebase-realtime-database
  • firestore
  • flask
  • flutter
  • google-cloud-video-intelligence
  • google-cloud-video-intelligence-api
  • google-gemini-api
  • google-maps
  • google-or-tools
  • huggingface
  • ifps
  • intelligence
  • jenkins
  • langchain
  • llamaindex
  • lora-fine-tuning
  • metamask
  • mist-model
  • mongodb
  • node.js
  • pinata
  • prometheus
  • pytorch
  • rag-(retrieval-augmented-generation)
  • react
  • rest-apis
Share this project:

Updates