Inspiration
1. Inspiration
The inspiration for Suraksha AI 2.0 arose from the critical shortcomings of current safety apps in the diverse and challenging Indian environment:
False Alarm Fatigue & Resource Misuse: Traditional SOS features either overwhelm emergency services (police/ambulance) with non-critical alerts or fail to send help when truly needed.
The Language Barrier: Existing apps often only support English, rendering them useless for the vast majority of non-English speaking citizens, especially in moments of panic.
Reactive, Not Proactive: Most apps only track location after a button is pressed, failing to prevent the incident or protect the user when they are incapacitated.
Digital Isolation & Systemic Vulnerability: A lack of specialized tools to protect highly vulnerable groups like the elderly from cognitive decline risks and the sophisticated fraud targeting them.
What it does
2. What it Does
Suraksha AI 2.0 transforms personal safety from a reactive notification system into a proactive, intelligent, and context-aware safety network.
Intelligent Triage: It instantly assesses the nature of the emergency using a critical question (in the user's local language like Telugu) and routes the alert to the most appropriate responder (Police for injury, Caregiver/Volunteer for non-violent/cognitive crises).
Silent Guardian: It automatically detects threats (struggle sounds, impacts) when the user cannot act, triggers a silent SOS, and secures encrypted evidence.
Dynamic Risk Mitigation: It provides predictive route safety based on real-time, micro-local environmental factors, not just historical data.
Role-Based Protection: It offers tailored security for specific users, such as monitoring routine deviation for elderly citizens.
How we built it
3. How we Built it
The architecture was designed for reliability, scalability, and deep localization:
Cross-Platform Core: Built using a versatile framework (e.g., React Native or Flutter) to ensure native performance and wide adoption across Android and iOS devices prevalent in India.
Localised AI Models: Utilized on-device (Edge) machine learning models (like TensorFlow Lite) for low-power, privacy-preserving Ambient Threat Detection (A-TD). This allowed for immediate classification of sounds (shouts, impacts) without sending raw audio data to the cloud.
Localization Layer: Implemented a robust localization framework with human-verified translations for all critical UI elements and, crucially, the AI Triage Prompt in multiple Indian regional languages (e.g., English, Telugu, Hindi, Marathi).
Fail-Safe Architecture: The SOS system was coded with dual-layer communication: primary (Data/Internet-based) and a Fail-Safe SMS Fallback to ensure alerts are sent even in low network areas, a common challenge in rural/remote India.
Secure Evidence Handling: Used end-to-end encryption for the audio/video evidence and a secure, tokenized API to share it with verified responders only.
Challenges we ran into
4. Challenges We Ran Into
Battery Management for A-TD: Ensuring the Ambient Threat Detection model ran continuously in the background without excessively draining the battery required extensive optimization of the on-device AI model's power consumption and frequency of sensor access.
Accurate Language Context: Implementing the AI Triage Question ('Is there injury?') across different Indian languages without losing critical context or appearing insensitive in a crisis was a significant hurdle, necessitating expert human translation and validation.
Synthesizing Dynamic Route Data: Since true IoT street-light data isn't widely available, creating a convincing "Micro-Local Dynamic Risk Score" required synthesizing real-time traffic, public event feeds, and simulated community reports to prove the predictive concept.
Privacy vs. Utility: Balancing the high-utility features like A-TD and Inactivity Monitoring with strict Privacy Guardrails (local processing of sensitive audio, explicit opt-in for monitoring) required careful design and transparent communication.
Accomplishments that we're proud of
5. Accomplishments That We're Proud Of
Achieving Intelligent Triage: Successfully developing a system that transitions from a simple button press to a context-aware decision-making engine that routes calls accurately, addressing the critical problem of resource misuse.
Deep Regional Localization: Achieving seamless, contextually accurate support for the Triage Question in Telugu (and other target languages), a feature that directly impacts the safety and efficacy for a huge, underserved user base.
Developing a Proactive Fail-Safe: Integrating the Ambient Threat Detection (A-TD), which provides genuine protection when the user is physically unable to press a button, marking a major step forward from conventional apps.
Focusing on the Forgotten: Successfully designing the Cognitive Load & Inactivity Monitor to provide a specialized layer of protection for the elderly, a demographic often overlooked by general safety apps.
What we learned
6. What We Learned
Context is King in Crisis: We learned that in an emergency, the nature of the crisis (e.g., injury vs. disorientation) is more important than just the location. The Triage system validated that even a two-choice question can dramatically improve response accuracy.
On-Device AI is Essential for Trust: For highly sensitive features like ambient audio detection, keeping the processing local on the user's phone is crucial for user adoption and trust, despite the technical complexity.
Localization is not just Translation: Designing for India requires thinking about the cultural perception of emergency responses and ensuring the visual design and language hierarchy cater to users with low digital literacy.
What's next for Suraksha AI 2.0
7. What's Next for Suraksha AI 2.0
Integration with Official Networks: Work with local police and emergency dispatch centers in major metros (starting with Hyderabad) to integrate the AI Triage Routing for genuine, verified alert channeling.
Enhanced Dynamic Data Sources: Integrate with true IoT endpoints (like public lighting sensors) and crowdsourced, verified data platforms to improve the accuracy of the Dynamic Route Safety Score.
Voice-Activated Triage: Develop an advanced local-language ASR (Automatic Speech Recognition) model to allow users to bypass the screen and state their injury/crisis verbally in Telugu or Hindi, enhancing hands-free emergency functionality.
Blockchain Evidence Tracking: Implement a secure ledger to provide a cryptographically verified and immutable chain of custody for all recorded evidence, ensuring it is immediately admissible and trusted by law enforcement.
Log in or sign up for Devpost to join the conversation.