EcoCell AI
Target User: Mobile Network Operators (MNOs) and Telecom Radio Access Network (RAN) Operations / NOC Engineering Teams managing dense 4G/5G urban deployments. What goes wrong today: 1. Massive Nocturnal Waste: Radio base stations consume 73% of total telecom energy OPEX, running full-power RF power amplifiers 24/7 even during nocturnal hours when subscriber traffic drops by -85%. 2. Rigid Vendor Timers Cause Congestion: Existing OEM static timers shut sectors on blind clocks, creating severe congestion (over 1,000 hours/year) during unexpected night traffic surges (venues, nightlife exits). 3. Critical SLA & 112 Safety Risk: Operators avoid sleep modes because waking deep-sleep cells takes too long, risking dropped emergency calls and carrier-grade SLA penalties.
Preliminary score
- Value for the user19.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI18.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing13.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility8.5 / 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 user is specific: RAN/NOC teams at mobile operators. The pain is concrete, with idle night-time radio power and congestion caused by static OEM timers, and you tie it to a clear money metric (AZN per year). Your testing goes beyond most submissions with a three-way comparison over 8,760 hours (always-on, static vendor timers, EcoCell) and an XGBoost vs Lag-24 MAE benchmark. You also report honest failures and fixes: the 30x power overestimate corrected to EARTH model values, and the 14 toggles/hour flapping fixed with hysteresis. Your dual-layer design, which keeps the macro umbrella always on while small cells sleep, is a sensible way to address the 112/SLA objection.
- To improve
- Your numbers don't agree across materials. The deck claims 480 t CO2 and 100k AZN for 850 sites, while the submission gives 58 t and 196k AZN for 500 sites. Words like 'audited', 'verified' and 'zero dropped calls' overstate what is a simulation on Kaggle data mapped to Baku coordinates. Make clear which results are simulated, show how the 9.2 ms wakeup and congestion hours were computed, and publish the forecaster's error on night-surge events specifically. You should also explain how EcoCell differs from vendors' existing AI energy-saving features (e.g. Ericsson/Huawei), and give a concrete pilot step with an operator: data access, O-RAN integration effort and running cost.
- GPT-6 AstraPreliminary62/100
- Strengths
- Your project targets a concrete operating expense for mobile operators and compares measurable savings against both always-on operation and static timers. Your reported power-calibration and switching-oscillation failures make the testing account specific and useful. Separating always-on macro coverage from sleeping capacity cells is a sensible design choice, and forecasting has a meaningful role in anticipating demand.
- To improve
- Your materials should show a reproducible telemetry-to-forecast-to-sleep decision trace, clearly separating precomputed simulation from actual network actuation. Publish the dataset identifier, chronological train/test split, baseline definitions, and scenario logs supporting the annual benchmark and failure fixes. Validate safety through measured added latency and overload/failover tests rather than treating macro coverage as proof of zero dropped calls or 0.0 ms latency. Specify operating and integration costs, reconcile the Mbps/MB/s and emissions figures, and define a realistic operator shadow-mode pilot.
The team
1. Iqbal Mirzayev
data
Omar BashirliLead
product
İsmayil Baba
developer
Süleyman Salmanli
designer