All projects
data
Process a stream and keep running totals
Windowed aggregation over an event stream, with late data handled honestly.
Why it matters
Batch is comfortable; streaming forces you to confront time, ordering and lateness. Those ideas keep reappearing in distributed systems interviews.
What you build
- Tumbling and sliding window aggregates
- Handling for events that arrive late
- State that survives a restart
- Correct results proven against a batch run
If you want more
- Add exactly-once output
- Add watermark tuning
Build it
- Decide what happens to events that arrive after their window closed
Shows you can reason about event time versus processing time, which is the core difficulty of streaming.
Get a plan built around projects like this