Systems under real constraints.

The problem

Teams ship quickly and discover security problems after release. Reviewers are inconsistent, context is scattered across tools, and remediation advice often arrives without enough context to act on.

The approach

• Orchestrate code review and vulnerability analysis through a single AI layer.
• Detect issues in context rather than as isolated scanner output.
• Generate remediation guidance that a developer can evaluate and apply.
• Keep the human reviewer in control of the final decision.

The problem

A single generic model struggles with the range of signals a real environment produces. Analysts drown in alerts that lack context, while response often happens in a different tool from detection.

The approach

• Split detection, analysis, and response across specialised agents.
• Let agents collaborate instead of duplicating the same analysis.
• Treat observability and scalability as design requirements, not afterthoughts.
• Keep the system explainable enough for a human analyst to trust.

The problem

Road damage is widespread, inspections are slow, and repair budgets are limited. The people who decide where to spend money often lack a current, spatial picture of where the damage is worst — and by the time a pothole is visible, the water has already done the structural damage.

The approach

• Detect potholes with a YOLOv8 computer-vision pipeline and an operator-facing review queue.
• Analyse satellite imagery to find macro-erosion patterns in areas that inspection vehicles cannot easily reach.
• Correlate road damage with topography and rainfall history to flag erosion incubation zones.
• Design the institutional layers — command link, contractor audit, aerial mapping — as documented extensions rather than claiming live government integrations.

System blueprint

Backend

Python · FastAPI · YOLOv8

Dashboard

Next.js 15 · TypeScript · Tailwind

Real-time

WebSockets for hazard events

Deployment

Turnkey setup · Ubuntu / WSL2 · RTX 3070+

System layers

• Pothole detection

  Frame-level vision inference that turns road footage into located, reviewable hazards.

  Working core

• Satellite sentinel

  Macro-scale raster analysis for erosion patterns across dense, hard-to-inspect areas.

  Working core

• Flood-erosion correlation

  A risk index combining elevation and rainfall history to predict where asphalt will fail.

  Working core

• Traffic command link

  A WebSocket channel for pushing diversion suggestions to traffic authorities.

  Designed extension

• Contractor audit ledger

  Before-and-after repair evidence reviewed before payment is released.

  Designed extension

• Coordinated aerial mapping

  A proposed multi-drone layer for covering all 20 Lagos LGAs.

  Designed extension

Layers marked “designed extension” are documented in the repository as intended architecture — not live government integrations.