The problem Suraksha solves

In it's current state, India lacks a nation wide threat analysis and social alert based system which exists for the benefit of the common individual.

Our application which is present as both, mobile and web-based apps aim to make it easy for individuals to report incidents of all sorts - motorvehicle accidents, natural disasters such as flood and earthquake, human conflicts such as riot, communal violence and war with the help of just a photo click with your mobile and the application does the rest for you.

It takes into account the geographical co-ordinates of the image which gets fed to a Google Gemini vision-to-text based ML model which generates the description of any activity in that image. The resultant text is fed to a sentiment analysis ML model which undertakes the calculation of the severity of the incident. This information is forwarded to the general public, law enforcement agencies and emergency services such as fire brigade and ambulances for further use.

Salient Features & Problems It Solves:

1. Mitigates the bystander effect: Reduces the hesitation in individuals to take action as all they have to do now is click a photo.
2. Reduces crucial time: It can result in help arriving quicker, hopefully saving more lives
3. One-click easy solution: Everything done with just the push of a button, literally.
4. Threat classificaton: Classification of threat level done helps in identifying potential hotzones to avoid.
5. Localized Alerts: Based on the user's current geolocation, the app will be able to send alerts of any potential safety hazard or incident in the neighbourhood.
6. Dedicated channel for monitoring incidents: Each type of emergency services can use our application as a dedicated channel for monitoring services unlike fragmented social media.
7. Better privacy features: To prevent defamation and privacy violations, the feed of the general public will have the faces of the individuals as censored.

Challenges we ran into

Some of the challenges which we faced while creating this project are mentioned below along with their solutions:

Moral/Legal Challenges:
Privacy Violations: Reported incidents can have individuals in the background of that image who might happen to be innocent or simply not want to be on public domain.
Solution #1: We have implemented a machine learning model which detects faces and blurs them whenever it goes to the feed of the public to safeguard their privacy. However, law enforcement agencies will retain the full right to utilize the images that don't contain any censorship to aid them in their investigation and other legal/search procedures.

Fake Alerts: The application can be utilized to plant fake alerts and cause mass hysteria.
Solution #1: Administrative right to law enforcement agencies to remove certain posts to safeguard the mental sanctity of the society.
Solution #2: (Not Implemented Yet - due to time constraints) Addition of a Deepfake detection model that detects fake/AI planted photos & model.
Solution #3: If multiple people keep reporting the same incident within the same vicinity, chances of it being severe and true rises.

Technical Challenges:

1. Lack of conflict/incident specific dataset.
   Sol: We made to make the dataset by scrapping relevant information surrounding conflicts and related incidents from social media posts and other news sources.

2. Dataloss during distribution on web app
   Sol: We had to make a redis queue to catch the posts of the users as usually there would be a lot of traffic resulting in data loss. After catching the posts, we distributed it along several workers which in turn greatly reduced the load across the platform increasing the efficiency and speed.