AI Engineering · Full-Stack · Geospatial

NAIMOS Intelligence

Decision-support prototype for monitoring and prioritizing suspected illegal-mining incidents in Ghana.

Status: Prototype
Year: 2026
Role: Designer and developer
Client: Independent project

Overview

A role-based platform with a PostGIS-backed API, an explainable 0 to 100 risk engine, an interactive map, evidence management and an AI assistant restricted to allowlisted queries. All data in the demo is synthetic and labeled as such.

Python
FastAPI
PostgreSQL / PostGIS
SQLAlchemy
Alembic
Next.js
TypeScript
MapLibre
Docker
pytest

The challenge

Field teams need to decide where to investigate first across many reports, with evidence that holds up and AI output that is clearly separated from verified fact.

Making AI output trustworthy enough for evidence work
Modeling geospatial data and an immutable evidence trail

System requirements

Ranked AI risk map answering where to investigate first, with week-over-week change from real score history
Incident lifecycle with a full status audit trail
Hashed, versioned evidence uploads and an AI-assembled evidence summary
Voice-to-field-report flow with GPS, photo and video capture
Predictive hotspot view that is explainable, capped below certainty, and paired with a patrol recommendation
Seven communication formats built only from database records, separating verified facts from AI interpretation

Solution architecture

Responsibility

Next.js web app

Role-specific dashboards, map, incident and evidence screens, mobile-friendly field capture.

Implementation

Overview of the public NAIMOS Intelligence PRO demonstration, carrying its own banner: simulated demonstration, incident records and figures are fictional.

Technical decisions

Never let the model write free-form SQL; only allowlisted, parametrized queries against a read-only role
Label every AI output and keep verified facts separate from interpretation
Count detected objects as the maximum in any single file, never the sum, to avoid double-counting
Keep the risk score rule-based so every point is explainable

Results and validation

Content needed: measured results with period and method. Never estimated.

Evidence

Public repository with architecture, database, AI guardrail and security documents
14 backend test files
Public simulated PRO demonstration

GitHub Live demo

Lessons learned

Content needed: what you would do differently

Content checklist (remove once filled)

Screenshots of the map, risk leaderboard and evidence package from the full app (the public demo shows only the PRO overview)
Whether any organization has reviewed or adopted it. The repo calls it a prototype with simulated data, so keep that wording
The satellite change-detection step is simulated in the repo; say so wherever it is described