AutoRisk AI: Automotive Risk Intelligence
Predict Risks. Protect Production. Strengthen the Supply Chain.
Inspiration
Automotive manufacturing depends on reliable suppliers, consistent production, and strong quality control. A supplier's quality issues, delivery problems, or repeated defects can create operational challenges, while variations in production output may signal areas that need further investigation.
We built AutoRisk AI to explore how data analytics can help automotive teams identify potential risks earlier and prioritize issues that deserve attention. Our goal is to transform operational data into understandable risk indicators that support better, data-informed decisions.
What It Does
AutoRisk AI analyzes automotive manufacturing and supply-chain data to highlight potential supplier and production risks.
Key features include:
- Supplier risk scoring: Combines quality ratings, on-time delivery performance, and defect indicators into a weighted risk score.
- Defect analysis: Identifies suppliers associated with critical and high-severity defects.
- Production monitoring: Compares average units produced, scrap quantities, downtime, and efficiency across production shifts.
- Risk categorization: Groups suppliers into high-, medium-, and low-risk categories to support prioritization.
- Data validation: Examines data ranges and missing values to help identify issues that could affect risk assessments.
- Dashboard concept: Presents risk indicators and supplier rankings in a format intended to make operational insights easier to explore.
How We Built It
We used Snowflake as the primary data platform and SQL to explore, aggregate, validate, and analyze automotive operational data.
The project uses data from several domains, including suppliers, purchase orders, quality defects, inspections, warranty claims, and production shifts.
We created analytical views to organize the data and support two main analyses:
- Supplier Risk Analysis: Supplier-level defect and purchase-order summaries are combined with quality and delivery indicators. A weighted, rule-based scoring method produces a risk score from 0 to 100.
- Production Risk Analysis: Production shift metrics are aggregated to compare output, scrap quantities, downtime, and efficiency across shifts.
The supplier scoring model uses provisional weights for quality risk, delivery risk, critical defects, and high-severity defects. Risk bands are assigned using defined thresholds. These rules provide an interpretable starting point and require further validation before real-world operational use.
The intended presentation layer is a Streamlit dashboard connected to Snowflake, allowing users to explore supplier risk scores, risk categories, and production insights.
Challenges We Faced
One major challenge was understanding the structure of the data across multiple schemas and identifying the correct relationships between suppliers, purchase orders, and defects.
We also needed to avoid double-counting records when combining supplier-level information. To address this, we aggregated defect and purchase-order data separately before joining the summaries.
Another challenge was validating the meaning and scale of the source data. For example, supplier quality ratings ranged from 50 to 99.5, which meant an initial threshold designed for a 0–5 scale was inappropriate. We also identified delivery percentages above 100 that require further investigation.
Finally, connecting the analytical views to the Streamlit application exposed a database-access issue in the application's runtime environment. This highlighted the importance of deployment permissions as well as analytical correctness.
What We Learned
Through this project, we learned how to use SQL views to organize analytical workflows, combine operational datasets, build transparent scoring rules, and interpret manufacturing metrics.
We also learned that a risk score is only as useful as the data and assumptions behind it. Validating source values, documenting scoring assumptions, and distinguishing an indicator from a proven prediction are essential when building decision-support tools.
What's Next
Future improvements include completing the live dashboard integration, validating scoring weights with domain experts, adding time-based trends, normalizing defect rates by supplier activity, and evaluating the scoring method against historical outcomes.
Our long-term goal is to develop AutoRisk AI into a more comprehensive automotive risk intelligence platform that helps teams investigate potential issues and make better-informed decisions.
AutoRisk AI turns operational data into clearer, more explainable risk insights for automotive manufacturing and supply chains.
Log in or sign up for Devpost to join the conversation.