USA Team
Our primary users are retail operations managers and fulfillment teams responsible for coordinating product deliveries and installations. In retail, a delivery being marked as complete does not always mean the customer's experience is complete. An order may arrive while installation remains unscheduled, or operational issues may put a promised delivery at risk. When teams lack a clear view of these risks, they may discover problems too late to prioritize preventive action. Operations teams need more than another analytics dashboard: they need to understand which orders require attention, why they are at risk, and what action to consider next. SYNOVA AI addresses this problem by combining order risk monitoring, explainable risk factors, experimental machine learning predictions, and intervention tracking in one operational workflow.
Preliminary score
- Value for the user16.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI15.5 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing12.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility6.5 / 15
Data requirements, running costs and a clear next step.
- Originality4.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary59/100
- Strengths
- Your focus on the gap between completed delivery and pending installation identifies a concrete problem for retail operations teams. Your testing report covers multiple workflows and documents specific fixes, especially correcting the overdue count from four to one. You also clearly distinguish experimental predictions from validated business outcomes and disclose the session-only intervention tracking.
- To improve
- Provide a reproducible order-level walkthrough showing an actual ML prediction, its explanation, and the resulting intervention; the supplied materials describe this flow without demonstrating its execution. Evaluate predictions on representative held-out data against simple rules, reporting false alerts and missed delays. Define a focused retail pilot with a baseline and adoption metric, such as time to identify and assign an at-risk order. Document ML-service setup, required operational data, estimated running costs, and the next step toward persistent intervention tracking.
- Claude Opus 5.5Preliminary50/100
- Strengths
- Your written submission narrows a broad deck into a concrete user: retail operations managers coordinating deliveries and installations. It names a real pain: orders marked delivered while installation is still unscheduled. Your testing write-up is unusually honest and specific: 96 automated tests, a found-and-fixed overdue-count bug (4 shown vs 1 actual), mobile layout fixes, and a documented session-only persistence limitation. You also flag clearly that the ML models are experimental and trained on synthetic data, which builds credibility.
- To improve
- The core risk view is rule-based, and the scikit-learn models are optional and reported with no metrics. Show precision/recall or ROC-AUC on a held-out split of your 1,000 synthetic records, and compare them against the rule-based score so we can see what the ML adds. The track asks for an adoption metric, so name one, such as the share of at-risk orders flagged before the promised date or hours saved per manager versus today's spreadsheet or manual follow-up. Add running costs, the data fields a retailer must supply, and a concrete pilot plan. Also align your deck, which spans four industries generically, with the focused retail story in your submission.
The team
Səma Abdullazadə
designer
Aydan ShahaliyevaLead
product
Ümüd Quliyev
data