# kafka-streams

Companion code for *Kafka Streams with Spring Boot 4: Real-Time Fraud Detection with Vector
Similarity* on [ankurm.com](https://ankurm.com). The whole topology runs under
`TopologyTestDriver`, so **no Kafka broker is needed** to run anything here.

| Stack | Version |
|---|---|
| JDK | 25 (Temurin 25.0.4.1+1) |
| Spring Boot | 4.1.1 |
| Spring for Apache Kafka | 4.1.1 |
| kafka-streams / kafka-streams-test-utils | 4.2.1 (what Boot 4.1.1 manages) |

## Run it

```bash
./scripts/run-all.sh        # runs the tests, rewrites output/*.txt
```

## What is where

| File | Role |
|---|---|
| `FraudTopology` | The topology as plain Kafka Streams code: similarity scoring, 5-minute velocity count, blocklist KTable join |
| `Embedder` | A hand-built 6-number feature vector and cosine similarity (not a learned model) |
| `FraudStreamsConfig` | The only Spring wiring: `@EnableKafkaStreams` plus handing the `StreamsBuilder` to the topology |
| `FraudTopologyTest` | Five tests; each one writes the matching `output/NN-*.txt` |
| `ContextLoadsTest` | Loads the Spring context with `auto-startup=false` to show the wiring without a broker |

## Captured output

Every file in `output/` is written by a test that also asserts it.

| File | Shows |
|---|---|
| `01-similarity-scores.txt` | Scores for five sample transactions against the 0.98 threshold |
| `02-similarity-alerts.txt` | Only the fraud-shaped transaction produces an alert |
| `03-velocity-window.txt` | Six transactions in two minutes give one alert; the next window starts from one |
| `04-blocklist-table-join.txt` | Join before blocking, after blocking, and after a tombstone |
| `05-topology-description.txt` | `Topology.describe()` of the whole graph |
| `06-spring-context.txt` | The Spring factory bean is wired and not running |

The article carries the explanations; there is deliberately no `docs/` folder.