Kavach

Project Overview

Kavach is a mobile-first personal safety companion engineered during the PromptWars x GDGoC MM(DU) Hackathon, where it secured a Top 30 position among 150+ student participants. Traditional panic button apps suffer from two major flaws: they require obvious manual activation at the exact moment a victim cannot safely operate their phone, and they merely alert remote contacts without offering any immediate actionable safe harbor. Kavach revolutionizes physical safety by combining a stealth incoming-call disguise with silent duress code detection, real-time routing to the nearest 3 verified SafePoints (pharmacies, security desks, warden posts), Google Gemini-powered multi-factor distress urgency scoring with a fail-safe sub-3s fallback, and passive journey anomaly monitoring.

System Architecture & Real-Time Engine

Kavach combines a responsive Next.js 14 App Router frontend with a high-throughput Node.js & Express backend. Real-time alert broadcast is powered by Socket.io, pushing live GPS coordinates, battery levels, and audio markers to guardians within 2 seconds. The data tier leverages PostgreSQL managed via Drizzle ORM, orchestrating relations for users, trusted contacts, emergency alerts, verified SafePoints, and passive journey breadcrumbs. All write operations are strictly validated against Google OAuth JWT session tokens to prevent unauthorized location spoofing.

AI Intelligence & Dual-Mode Urgency Scoring

Distress severity is evaluated using Google Gemini models via the Vercel AI SDK (`generateObject`) structured by Zod schemas. The AI analyzes free-text distress messages, ambient background context, time-of-day risks, and proximity to known safety havens to assign an Urgency Score from 0 to 100. • **Fail-Safe Fallback:** Because emergency tooling cannot depend on external LLM uptime or network spikes, the AI pipeline races against an internal 2.5-second timeout. If the LLM call times out or errors, the system automatically falls back to deterministic rule-based lexical analysis, guaranteeing sub-3s scoring without blocking the SOS dispatch.

Stealth Protection: Fake Call & Silent Duress

For situations where pressing an obvious SOS button would escalate an attack, Kavach provides a deceptive incoming phone call screen. The user can interact with the disguised call interface naturally while speaking or typing a pre-configured silent duress keyword. The background service silently registers the trigger, captures pinpoint GPS coordinates, and alerts guardians with zero visual flash or UI change.

SafePoint Routing & Guardian Live Map

Rather than leaving someone in danger waiting helplessly for help, Kavach automatically calculates walking directions to the 3 closest verified safe points (24/7 pharmacies, campus security desks, warden offices, and police kiosks) using Leaflet and OpenStreetMap tiles. Simultaneously, trusted contacts receive a real-time guardian dashboard with synchronized live map tracking, battery metrics, and one-tap emergency escalation.

Journey Shield: Passive Anomaly Detection

Journey Shield enables hands-free passive travel monitoring for night commutes and solo transit. The background worker continuously tracks GPS pings against planned routes. If the traveler stalls unexpectedly, deviates off-course, or fails to confirm safe arrival within the expected ETA window, Kavach initiates automated check-in alerts and auto-escalates to trusted guardians.

Engineering Challenges & Solutions