An interview-ready operational analytics portfolio project built from real U.S. airline performance data. The application converts hundreds of thousands of flight records into governed KPIs, regional scorecards, interactive diagnostics, and plain-language explanations.
Operational teams often spend time moving between dashboards, applying filters, and manually tracing changes. This project demonstrates a more direct workflow: one Python pipeline prepares reusable metrics and structured JSON, while an interactive dashboard and diagnostic copilot use the same analytical foundation.
An example question is: “Why did on-time performance decline this week?” The copilot compares the selected period with the previous week, identifies the largest delay category, and highlights the airport and carrier segments that deserve investigation.
- Python data pipelines and reusable metric definitions
- Ten operational KPIs across national and regional views
- Weekly trend, carrier, airport, and delay-cause analysis
- Structured JSON output that supports multiple user experiences
- Interactive HTML dashboard without a server or external JavaScript libraries
- Plain-language diagnostic explanations grounded in displayed metrics
- Responsible decision support: the interface separates signals from confirmed causes
flowchart LR
A[U.S. DOT/BTS flight records] --> B[Python streaming pipeline]
B --> C[Validated KPI aggregations]
C --> D[Structured JSON]
D --> E[Executive scorecard]
D --> F[Diagnostic copilot]
E --> G[Operational decision]
F --> G
- Flight volume
- On-time arrival rate
- Average arrival delay
- Average departure delay
- Cancellation rate
- Diversion rate
- Average taxi-out time
- Average taxi-in time
- Carrier-delay share
- Weather-delay share
The committed dashboard already contains an aggregated July 2026 dataset. No setup is required:
python3 -m http.server 8000Open http://localhost:8000.
To rebuild the data from the official BTS archive:
python3 build_data.py --downloadRaw flight records are downloaded locally and excluded from version control. Only compact aggregated scorecard data is committed.
The project uses the Marketing Carrier On-Time Performance dataset from the U.S. Department of Transportation, Bureau of Transportation Statistics. A flight is considered on time when it arrives within 15 minutes of its scheduled arrival time.
- Source: BTS TranStats
- Demonstration period: July 2026
- Geographic grouping: U.S. Census-style regions based on origin state
- Unit of analysis: scheduled domestic flight
- Canceled and diverted flights are excluded from completed-flight delay averages.
- Cancellation and diversion rates use all scheduled flights as their denominator.
- Delay shares use BTS-attributed delay minutes.
- The copilot uses deterministic calculations so every statement can be traced to the scorecard data.
- The tool identifies patterns for investigation; it does not claim that correlation proves operational causation.
- Start with the national scorecard and compare the latest week with the previous week.
- Select What changed? to summarize direction, scale, and the leading delay category.
- Switch to a region to show reusable metric definitions across organizational dimensions.
- Select What drove performance? to connect the KPI movement with carrier and airport segments.
- Explain how the same JSON foundation could power Tableau, an internal agent, a custom GPT, or an API.
Asteway Kebede — Data Analytics & Applied AI
LinkedIn · GitHub