AI Patterns

Pipeline / Chain

Pipes and Filters

Process a message through a sequence of steps.

Fan-out / Broadcast

Multicast

Send the same message to multiple destinations in parallel.

Fan-in / Reduce / Join

Aggregator

Combine multiple related messages into a single message.

Scatter-Gather

Scatter-Gather

Send a message to multiple recipients and aggregate their replies.

Map / Transform

Transform, Set Body

Transform or replace the message content.

Split / Chunk

Splitter

Break a message into smaller pieces for individual processing.

Merge

Aggregator

Merge multiple messages into one using a correlation key and aggregation strategy.

Filter

Message Filter

Discard messages that do not match a predicate.

Router / Dispatch

Content-Based Router, Dynamic Router

Route messages to different destinations based on content or rules.

Enrich / Hydrate / Augment

Enrich, Poll Enrich

Fetch additional data from an external source and merge it into the message.

Validate / Guard

Validate

Check that a message satisfies a condition before further processing.

Normalize

Normalizer

Convert messages from different formats into a single canonical format.

Serialize / Deserialize

Marshal, Unmarshal

Convert between objects and wire formats (JSON, XML, Avro, Protobuf, etc.).

Retry

Error Handler

Automatically retry failed message processing with configurable backoff.

Circuit Breaker

Circuit Breaker

Stop calling a failing service and provide a fallback response.

Fallback

On Exception, On Fallback

Handle exceptions with onException (general), or define a Circuit Breaker fallback with onFallback.

Dead Letter / Poison Pill

Dead Letter Channel

Move messages that cannot be processed to a separate channel for investigation.

Compensate / Rollback

Saga, Rollback

Undo or compensate for completed steps when a multi-step process fails.

Rate Limit / Throttle

Throttler

Limit the number of messages processed per time period.

Debounce / Delay

Delayer

Introduce a delay before processing a message.

Poll / Pull

Polling Consumer, Poll

Consume messages on demand rather than being pushed.

Loop / Iterate

Loop

Repeat processing a message a fixed number of times or until a condition is met.

Sample

Sampling

Pick one message out of many in a time period to reduce volume.

Load Balance

Load Balancer

Distribute messages across multiple endpoints.

Round Robin

Round Robin Load Balancer

Distribute messages evenly in rotation.

Failover

Failover Load Balancer

Try the next endpoint when the current one fails.

Weighted

Weighted Load Balancer

Distribute messages according to weighted ratios.

Sticky / Affinity

Sticky Load Balancer

Route related messages to the same endpoint based on a correlation key.

Dedup / Deduplicate

Idempotent Consumer

Detect and discard duplicate messages based on a unique identifier.

Claim Check / Stash

Claim Check

Temporarily store message data and retrieve it later to reduce payload size.

Source / Ingress

From

Define the input endpoint that feeds messages into a route.

Sink / Egress

To, To D

Send messages to a fixed or dynamically computed endpoint.

Connector / Adapter

Channel Adapter

Connect an application to a messaging system.

Webhook / Event-Driven

Event Driven Consumer

React to messages as they arrive without polling.

Gateway

Messaging Gateway

Encapsulate access to the messaging system behind a clean interface.

Wire Tap / Tap / Observe

Wire Tap

Send a copy of the message to a secondary destination for monitoring.

Agent-to-Agent / A2A

A2A

Google’s Agent-to-Agent protocol for communication between AI agents.

Tokenize / Chunk (for LLM)

LangChain4j Tokenizer

Split text into tokens or chunks sized for LLM context windows, using LangChain4j tokenizer strategies.