SafeRaste

SafeRaste is a mobile-first Progressive Web App (PWA) designed to provide dynamic, time-aware route safety insights for women. It uses Cognee's full memory lifecycle — remember(), recall(), improve(), and forget() — to build a knowledge graph of crowdsourced safety reports and deliver live, time-aware safety scores for urban street segments.

Key Features

• Interactive Safety Map: A responsive, full-screen Leaflet map focused on Pune, India, featuring accurate topological road traces powered by OSRM.
• Time-Aware Insights: Safety scores change dynamically based on the selected time of day (Morning, Afternoon, Evening, Night). The same street can be safe at 2 PM and unsafe at 10 PM.
• Cognee-Powered Memory Backend: A FastAPI backend connects the frontend to Cognee's memory lifecycle APIs in real time — every report submission, safety query, and decay operation hits the actual knowledge graph.
• Temporal Decay (forget): Reports older than 60 days lose weight; reports older than 90 days are pruned entirely. Stale safety data is actively dangerous — this ensures the map reflects current reality.
• Contradiction Detection (improve): When conflicting reports exist for the same segment and time bucket, the system surfaces both signals honestly instead of false-averaging into a misleading score.
• "Light Left for You": An anonymous encouragement system where women leave short human notes for the next woman walking that route.

Technology Stack

• Frontend: React, Vite, Leaflet, Vanilla CSS
• Backend API: Python FastAPI hosted on Vercel Serverless Functions
• Memory Layer: Cognee Cloud Knowledge Graph (via lightweight HTTPX REST integration to bypass Vercel serverless size limits)
• Routing Data: OpenStreetMap Routing (OSRM)

Architecture

graph TD
    A[React / Vite Frontend] -->|HTTP POST| B[Vercel Serverless Function <br> fastapi api/index.py]
    B -->|HTTPX REST API| C{Cognee Cloud}
    
    C -->|/api/v1/remember| D[(Knowledge Graph)]
    C -->|/api/v1/recall| D
    
    D -.-> E[StreetSegment Node]
    D -.-> F[Report Node]
    D -.-> G[TimeOfDayBucket Node]
    
    E --- F
    F --- G
    
    classDef frontend fill:#3b82f6,stroke:#1d4ed8,stroke-width:2px,color:#fff;
    classDef serverless fill:#8b5cf6,stroke:#6d28d9,stroke-width:2px,color:#fff;
    classDef cognee fill:#10b981,stroke:#047857,stroke-width:2px,color:#fff;
    classDef graphdb fill:#f59e0b,stroke:#b45309,stroke-width:2px,color:#fff;
    
    class A frontend;
    class B serverless;
    class C cognee;
    class D graphdb;

Loading

Getting Started

Follow these instructions to run SafeRaste locally on your machine.

Prerequisites

• Node.js (v18 or higher recommended)
• Python 3.10+
• A Cognee Cloud account (free Developer plan with code COGNEE-35)

Installation & Setup

1. Clone the repository:

   git clone https://github.com/Professor471/SafeRaste.git
   cd SafeRaste

2. Set up environment variables:

   cp .env.example .env
   # Edit .env and add your Cognee API key and LLM provider key

3. Install Python dependencies and seed the knowledge graph:

   pip install -r requirements.txt
   python seed_data.py

4. Start the backend API server:

   uvicorn backend.main:app --reload --port 8000

5. Install frontend dependencies and start the dev server (in a new terminal):

   npm install
   npm run dev

6. View the application: Open your browser and navigate to http://localhost:3000. The Vite dev server proxies /api/* requests to the FastAPI backend automatically.

Project Structure

• src/App.jsx — The core React application: interactive map, time-of-day filters, report logging UI, and Cognee memory lifecycle console.
• src/App.css — Dark-mode optimized styles and responsive layout definitions.
• backend/main.py — FastAPI server exposing Cognee's remember(), recall(), improve(), and forget() as HTTP endpoints.
• seed_data.py — Synthesizes and ingests 120 route safety reports across 10 Pune street segments into the Cognee knowledge graph.
• test_memory.py — Phase 1 integration test: proves time-aware recall() returns different results for the same street at different times.
• test_phase2.py — Phase 2 integration test: demonstrates improve() confidence reweighting and forget() temporal decay.
• SafeRaste_Build_Process.md — Historical build logs and design rationale from earlier development phases.

How the Cognee Memory Lifecycle Powers SafeRaste

Cognee API	What SafeRaste Does With It
remember()	Ingests user-submitted safety reports as structured text into the knowledge graph, creating StreetSegment, Report, and TimeOfDayBucket nodes with typed edges.
recall()	Queries the knowledge graph for safety reports relevant to a specific street segment and time-of-day bucket. Returns evidence-backed safety assessments.
improve()	Runs post-ingestion enrichment. When multiple reports agree, confidence increases. When reports contradict, both signals are surfaced honestly.
forget()	Prunes stale datasets. Reports decay linearly after 60 days and are fully forgotten after 90 days, ensuring the map reflects current ground truth.

AI Assistance Disclosure

This project was built with assistance from AI tools, including Claude (Anthropic) for planning, architecture, and code generation, and Antigravity/Gemini for code review and auditing. All core design decisions and final implementation choices were made by the team.