confidence
From 0.0 to 1.0. How far to trust the level.
How it works
Evidence, not reassurance.
Nayki turns real-world signals into a risk level you can read, with the reasons beside it. This page explains the rules the backend follows and the shape of the system.
The risk model
Four levels. No "safe".
Every route and area gets one of these, plus the numbers that explain it.
- lower_riskThe evidence points to fewer concerns. It is not a promise.
- moderate_riskSome concerns show up in the evidence.
- higher_riskSeveral concerns show up, or a strong one does.
- unknownThere is not enough evidence to say anything.
Fig. 3 How one route's level is built. Four signals fill the bars and the level and a confidence value follow. With no data at all the result is unknown, never safe. The values shown are examples, not real data.
Every answer carries its evidence
Four numbers and a list of reasons.
evidence_count
How many records stand behind the answer.
coverage_status
unavailable, limited, moderate or strong.
risk_reasons
The named reasons behind the level, from a fixed list.
- poor_lighting
- recent_incident
- isolated_road
- limited_data
- help_far_away
- night_time
- government_data
- user_reports_unverified
When no evidence exists, the backend must return coverage_status: "unavailable", low confidence and a null score. It must never invent a high-safety claim to fill the gap.
Architecture
An app, an API, a spatial database.
- Android app
Kotlin and Jetpack Compose. Search, compare routes, see help nearby, report, trigger SOS.
- FastAPI service
Python 3.12, Pydantic v2, SQLAlchemy 2. Auth, routes, places, safety, incidents, SOS and privacy endpoints.
- Postgres with PostGIS, and Redis
Spatial data on an H3 grid at resolution 9, with Redis for caching.
- Outside sources
Clients for Google Places and Routes, GraphHopper, OpenStreetMap Overpass and Firebase push.
Privacy rules in the code
Built to know less about you.
No coordinates in logs
Exact latitude and longitude are stripped by log processors. Only H3 cells or general regions may be logged.
Hashed audit trail
Client IP addresses are hashed with SHA-256 before they reach the audit table.
Aggregated nearby data
Nearby incidents are grouped onto H3 cells, so people who report a hazard are not pinpointed.
Export and delete
Endpoints for exporting your data and deleting your account, with device tokens revoked.
API surface
What the backend exposes.
| Area | Path | Purpose |
|---|---|---|
| Health | /health | Service status. |
| Auth | /auth | Sign-in with Firebase tokens. A mock bypass exists for local development only. |
| Profile | /me | Profile, emergency contacts and devices. |
| Places | /places/search, /places/help | Destination search and nearby help. |
| Routes | /routes | Evaluate and compare routes. |
| Safety | /safety | Risk analysis and aggregated nearby data. |
| Reports | /reports | Crowd-sourced incident reports. |
| SOS | /sos | Create, cancel, respond to and close alerts. |
| Privacy | /privacy | Data export and account deletion. |