SignalFlow
TARGET USER: Municipal Intelligent Transport Systems (ITS) engineers, urban traffic management operators (e.g., Baku Transport Agency / BTA), and emergency dispatch coordinators managing high-density urban corridors. THE PROBLEM TODAY: Static Cycle Inefficiency: Fixed-timer traffic signals operate on pre-programmed intervals, creating artificial congestion when empty lanes hold green signals while high-density approaches sit idling. Economic & Environmental Cost: Idling vehicles consume millions of liters of wasted fuel annually, significantly inflating urban carbon emissions and commute delays along bottleneck arterial roads. Emergency Vehicle Delays: Critical emergency services (ambulances, fire engines) face life-threatening response delays due to an inability to dynamically pre-empt red lights at locked intersections. Lack of Adaptive Real-Time Infrastructure: Traditional road-loop sensors are expensive to install, maintenance-heavy, and lack vehicle classification intelligence (e.g., distinguishing between a sedan, bus, or emergency unit).
Preliminary score
- Value for the user19.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI20.5 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing18.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility7.5 / 15
Data requirements, running costs and a clear next step.
- Originality5.5 / 10
A meaningful difference from common solutions.
What the judges wrote
- Claude Opus 5.5Preliminary68/100
- Strengths
- Your testing is the standout: 180 automated safety tests plus 66 headless simulations comparing fixed timers, Webster, max-pressure and Gemini across 3 seeds and 4 demand levels. You honestly report where Gemini loses to Webster or max-pressure, along with the simulator bugs you found. Your layered architecture is well designed: Gemini proposes plans, max-pressure adapts every second, and a hard safety layer enforces yellow, all-red, minimum green and maximum red regardless of AI output. The side-by-side demo, with identical seeds, demand and a surge button, makes the comparison easy to see and fair.
- To improve
- Your own results show plain max-pressure beating Gemini at rush hour and Webster beating it at high demand, so you need to show more clearly what the LLM uniquely adds. That could be operator-readable explanations, handling unusual events, or emergency pre-emption, which you name in the problem statement but do not appear to implement. Feasibility needs work: estimate per-intersection API costs at real call rates, say how real detector or camera data from an agency like BTA would feed the system, and define a concrete pilot step such as calibrating on one real Baku corridor. Also remove the live Gemini API key from your submission text and rotate it.
- GPT-6 AstraPreliminary73/100
- Strengths
- Your side-by-side simulator uses matched demand and tracks concrete operational outcomes, including wait, queues and throughput. Your documented safety layer, bounded API usage and adaptive fallback make the architecture more credible than unrestricted LLM signal control. Your testing provides useful controller comparisons and specific failure cases, including conditions where Gemini does not outperform simpler methods.
- To improve
- Your next evaluation should use more seeds, longer runs and calibrated traffic traces, isolating Gemini's added value over Webster and max-pressure rather than only fixed timers. Your deployment plan needs a real queue-data source, signal-controller integration, a supervised pilot and estimated operating costs at city scale. Your emergency-priority use case remains unshown; implement and test it or narrow the stated scope. Your submitted API credential should be revoked and replaced with a server-side secret that is not distributed with the project.
The team
Farid Mehdiyev
developer
Riad Aghayev
designer
Gismat SuleymanovLead
product