Qube Research and Technologies
London
During my placement at QRT, I worked on a real-time dashboard that presents account information across funds and regions. My main project was to create a recorder and replay capability so engineers could capture the ordered stream of messages from the live system and investigate the same behaviour offline. Throughout the development I started with the prototype that has unresolved latency and overly specific implementation. I then redesigned the recorder as a generic, datasource-agnostic subsystem and delivered it into UAT. The replay implementation developed substantially giving us fine-grained control to step through the timeline and will soon be deployed separately. The placement strengthened my interest in systems engineering, performance analysis and production tooling.
My role was to take ownership of the recorder/replay project from architecture through implementation, testing, review and UAT enablement. I identified that the original design would not scale as more dashboard caches were added, and replaced it with generic datasource recording components while preserving the original live path when recording is disabled. I designed the persistence and lifecycle model, split the work into reviewable pull requests. Also, to improve efficiency, I analysed a six-day recording containing 94 millions payload frames and found that 97% were duplicates. I then ran a sizing experiment and implemented a content-addressed deduplication index with optimal capacity. When review exposed an exception-safety defect, I fixed it, added an additional regression test and documented the interface precondition. I also found an existing bug in the system and took ownership over it and ship a fix.
I used C#/.NET 9, asynchronous and concurrent programming, dependency injection, lifecycle management, append-only persistence, background processing, compression, hashing and content-addressed deduplication. The recorder was designed to keep heavy work away from a latency-sensitive callback path, and used sequencing and a materialisation frontier to reason about correctness for resuming. I strengthened my testing skills by building test methods, adding exception-path coverage and removing timezone assumptions from timeline tests. For performance analysis I used long-running runtime-counter sampling, threshold-triggered GC dumps and workload forensics over large recording artifacts. I also ran a parameter-sweep experiment and selecting a cache size from the measured knee.
I developed ownership, planning, communication, professional judgment and inclusive engineering. I owned this recorder project and convert it into small, testable deliverables that can be reviewed and tested incrementally and be deployed. I also responded selectively to feedback, accepting findings that improved safety and explaining why other suggestions would create inappropriate coupling. I also learned to direct the problem and issue to the corespondent quicker to keep the progress going. My inclusive engineering practice focused on reducing hidden assumptions and knowledge barriers by proper documentation and intuitive organization of the developed system.