N-PBS is Navblue's preferential bidding system, now part of Skywise. Crew bid for the trips and days off they want. The Scheduler builds each person's month by seniority while meeting legality rules and the airline's coverage targets. It schedules crew at major North American carriers.
PairingsMulti-day trips to cover
Crew bidsLikes, dislikes, days off
SchedulerC++ search engine, by seniority
AwardsEach crew member's month, with reasons
Simplified illustration. No proprietary details.
1
Extending a mature engine
The award logic, search operators, database layer and nervousness handling were there before me. I maintain them and add new capabilities in C++, on a product used by 35+ airlines.
2
One small change, every schedule
In seniority bidding, one tweak near the top can ripple down to everyone below. I traced how a single parameter change cascades through the awards, the problem the industry calls scheduler nervousness.
3
Searching smarter
A month of trips for thousands of crew is too big to solve by brute force. I tuned the large neighborhood search operators and added memoization, so the engine stops recomputing work it has already done.
4
The whole product, not just the engine
C++ and SQL in the core, Python and Perl tooling, an Angular front end, served on Apache and Linux, with build artifacts in JFrog.
Teams lost time hunting through wikis. I built an internal assistant on AWS that answers from Confluence and Google Docs, and teams use it.
DocsConfluence, Google Docs
ChunkAmazon EC2, cost-tuned
EmbedAWS Bedrock
StoreVectors in Amazon S3
AnswerRAG with context
Next: hybrid keyword and vector retrieval, reranking, and native S3 Vectors indexes, each change measured with rag-eval-harness.
