Urban traffic congestion is not merely an inconvenience — it is a measurable drag on economic productivity, air quality, and quality of life. As cities grow faster than their infrastructure, the answer is no longer wider roads. It is smarter ones. Profecia Links is at the forefront of deploying AI systems that make existing infrastructure work harder, respond faster, and learn continuously.
The Problem With Legacy Traffic Systems
Most traffic management systems in operation today were engineered for a different era. Fixed signal cycles, periodic manual reviews, and reactive incident response create a fundamental mismatch with the fluid, unpredictable nature of modern urban traffic. Operators are asked to manage complex, real-time systems with tools that produce data in hindsight.
The results are familiar to any city-dweller: signal phases that favour phantom queues; incidents that go undetected until gridlock has already propagated; and planning decisions made on weekly averages rather than minute-by-minute reality.
The technology to change this exists today. What has been missing is the architecture to apply it at the operational level — close to the hardware, in real time, with no dependence on cloud connectivity. That is precisely what Profecia Links builds.
Why AI changes the equation
Traditional signal controllers respond to presence; AI systems predict demand. The difference is the gap between turning the lights on when someone walks in, versus turning them on because you know they're about to. Prediction unlocks optimisation that reactivity cannot.
Profecia Links at Work: Wadi Saqra, Amman
Profecia Links is currently executing a Proof of Concept (PoC) for the Greater Amman Municipality (GAM) at the Wadi Saqra intersection — one of Amman's most strategically significant traffic nodes. The project demonstrates, in a live operational environment, two complementary AI capabilities that define our approach to intelligent traffic management.
Greater Amman Municipality — Wadi Saqra PoC
A 16-week proof of concept deploying two AI modules on GAM's existing data infrastructure — fully on-premise, zero cloud dependency, and engineered to meet GAM's data sovereignty requirements from day one.
16wk
Kick-off to UAT sign-off
≥85%
Incident detection F1 target
≤10%
Forecast error (MAPE) target
500ms
Frame-to-alert latency ceiling
Module 1 — Real-Time Incident Detection
The first AI module processes live CCTV streams using a fine-tuned YOLOv8 computer vision model, adapted for Jordanian road conditions and local vehicle types. Every frame is analysed in under half a second. A DeepSORT multi-object tracker maintains persistent object identities across frames — distinguishing a genuine incident from a momentary detection artefact — while a custom anomaly classifier grades each event by severity: Low, Medium, or High.
The system detects stalled vehicles, congestion spill-back, wrong-way driving, and pedestrian intrusions. Alert payloads — including a timestamp, camera ID, bounding box coordinates, severity score, and a video thumbnail — are emitted instantly to the operator dashboard and logged permanently for audit.
Module 2 — Flow Forecasting & Signal Optimisation
The second module is concerned with what happens next rather than what is happening now. An LSTM sequence model, trained on over twelve months of SCATS detector history, forecasts vehicle volume at the 15, 30, and 60-minute horizons for each approach arm of the intersection. A Temporal Fusion Transformer runs in parallel to cross-validate predictions and provide uncertainty estimates.
These forecasts feed a signal optimisation engine — built on SciPy and Google OR-Tools — that translates demand projections into concrete cycle length and phase split recommendations. Critically, the system is advisory: it surfaces data-driven recommendations to operators who retain full control of signal actuation. The model retrains nightly on a rolling 90-day window, autonomously correcting for seasonal drift without human intervention.
The goal is not to replace operators — it is to give them superhuman situational awareness and forecasts no human team could generate alone.
— Profecia Links, GAM Solution Architecture, 2026
Architecture That Respects Reality
Smart city deployments often founder not on algorithmic sophistication but on operational constraints ignored at design time: intermittent connectivity, data sovereignty law, legacy hardware, and procurement timelines that lag behind project schedules. Profecia Links engineers for these constraints from the first line of architecture.
100% On-Premise · Air-Gapped Operation
The GAM system operates entirely within the municipality's own data centre. No video frame, no detector count, and no operator decision ever leaves the GAM network. After initial offline mirroring of container images and packages, the system has zero internet dependency. This is not a configuration option — it is a first-class architectural requirement.
Containerised Modularity
The system is structured as eleven discrete Docker Compose services — from video ingest and GPU inference to database storage and the operator dashboard — communicating over internal message buses (Redis Streams). Each module can be developed, tested, validated, and updated independently. Failures are contained; updates are surgical.
Technology Stack
Every component is open-source with no commercial licence fees, reducing long-term operational cost and vendor lock-in.
YOLOv8
DeepSORT
PyTorch 2.x
LSTM / TFT
FastAPI
PostgreSQL 15
TimescaleDB
Redis Streams
Grafana 10
OR-Tools
Docker Compose
CUDA 12 / TensorRT
Designed to Scale
The Wadi Saqra PoC targets one intersection. The architecture targets a city. Adding a second intersection requires only a new CCTV ingest configuration and an additional inference container stack — the database, dashboard, and API layers need no changes. Profecia Links' estimate is two to three weeks of integration engineering per additional intersection after the PoC, giving GAM a clear and costed path to city-wide rollout.
Measuring What Matters
Profecia Links works to measurable outcomes, not intentions. Every PoC engagement is governed by a formal KPI framework with independently verifiable measurement methods. For GAM, eight acceptance KPIs define success:
| KPI | Target |
|---|---|
| Incident detection F1 score | ≥ 85% |
| Incident false positive rate | ≤ 5% |
| Traffic volume forecast MAPE (30-min) | ≤ 10% |
| Frame-to-alert latency (P95) | ≤ 500 ms |
| Dashboard alert delivery latency | ≤ 2 seconds |
| System uptime (evaluation period) | ≥ 99% |
| Air-gap compliance | 100% |
| Operator usability score (UAT) | ≥ 4.0 / 5.0 |
The Wider Picture: AI and the Future of Smart Cities
The Wadi Saqra deployment is a single intersection in a single city. But the principles it embodies are universal. Every city in the world has intersections where traffic flow is suboptimal, where incidents go undetected for too long, and where signal timing is set by schedule rather than demand. The systemic opportunity is vast.
What differentiates Profecia Links' approach is the recognition that AI in public infrastructure must earn institutional trust before it earns operational authority. The GAM system is advisory by design — it presents optimised recommendations, it does not impose them. Operators remain in the loop. Over time, as the system demonstrates its accuracy across seasons and conditions, the scope of automation can expand at a pace the institution is comfortable with.
This staged autonomy model is not a limitation. It is the only responsible architecture for critical urban infrastructure.
From PoC to city-wide platform
The containerised architecture of the Wadi Saqra system was designed from day one to scale. Adding a new intersection to the platform requires provisioning additional ingest capacity and a new inference container stack — the intelligence layer, the database, and the operator dashboard scale horizontally with minimal re-engineering. Profecia Links estimates 2–3 weeks of integration work per additional node.
Why Profecia Links
Profecia Links brings together the disciplines that large-scale AI infrastructure demands: computer vision engineering, ML operations, system integration, and deep domain experience in urban mobility. We design solutions that survive contact with real operational environments — with legacy hardware, with data sovereignty constraints, with teams who need tools that enhance rather than replace their expertise.
Our engagement model is outcomes-first. We define measurable KPIs before the first line of code is written. We deliver working systems against those KPIs. And we design every system to be maintained, extended, and trusted by the teams who operate it.
The GAM partnership is a demonstration of what is possible when technical ambition is matched with operational rigour. We are proud to be part of Amman's journey toward a smarter, more responsive city.
Ready to bring AI to your traffic infrastructure?
Our team works with municipalities and transport authorities to define, build, and validate AI systems that deliver measurable outcomes from day one.
