event-time

Pulse is a tiny real-time data streaming framework (mini Flink/Beam) in Rust. Async (Tokio), pluggable operators, state, and I/O. Fast, modular, local-first.

A stream processing engine that stays correct when processes die. SQL or typed C++ over event time and keyed state, with exactly-once delivery and twenty-plus connectors: brokers, CDC, databases, lakehouse tables, graphs. The same pipeline runs on a single machine, inside a service or scaled out across a highly available cluster. Arrow-native C++.

Deterministic reconciliation engine with typed breaks, replayable decisions, and audit trail.

Delayed conversion labels: freshness, correctness, corrections, processing lag, and retained state.

Real-time decisions on an electricity distribution network. Deterministic code owns whether a window is closed; models only forecast, score and rank. Seven claims proved offline and against a deployed AWS estate — 0 rows published early of 3,779, 41 gate mutations refused, all six erasure legs confirmed.

Event-time windows and watermarks for Ruby. AMS Akidau, NCC-4219.

Real-time ML features from a click stream with event-time windows, where late and out-of-order events are handled explicitly; a reference engine defines the right answer and the PyFlink job is checked against it.

Streaming telemetry for AI agents: Kafka → PyFlink event-time windows (late, out-of-order events) → per-minute latency and tool-call metrics, enriched from Postgres, on a live dashboard.

Cross-engine output validation for event-time streaming pipelines.

Interactive F# lab for watermark lag, late events, idle partitions, incomplete windows, and delayed alerts.

Real-time anomaly detection on payment streams with PyFlink: event-time windows, keyed state, watermarks

Kafka streaming lab demonstrating partitioned topics, duplicate handling, late events, DLQ routing, and measured end-to-end latency.

MoonBit event-time correction engine: atomic retractions, session merge/split, finality guards, downstream changelogs and replay checkpoints.

Exactly-once streaming pipeline: Kafka KRaft to Flink to ClickHouse. Zero data loss verified by killing Flink mid-stream, not by assertion.

Transactional event-time windows with durable source receipts, immutable outbox outputs, process-crash recovery tests, and lateness tradeoff experiments.

Kafka -> Spark Structured Streaming -> Delta, with event-time windowing and watermarking made runnable: a dependency-free reference implementation you can execute without a JVM

Kafka-compatible streaming demo for real-time birding intelligence: event-time windows, late-event handling, stateful deduplication, hotspot metrics, and target-species alerts.

Watermark Lens - diagnostic tooling for Apache Flink watermarks, windows, and late data

Stream processing in Rust with a Flink-inspired DataStream API and runnable benchmark workloads.