Kroxylicious · Merged
Avro and Protobuf validation on the Kafka write path ↗
Authored schema-validation support for the Record Validation filter, bringing additional schema formats into proxy-side enforcement.
ImplementationWork & open source
Core infrastructure, upstream fixes and focused experiments. Each project starts with an engineering question and makes the answer inspectable.
Upstream contributions
Kroxylicious · Merged
Authored schema-validation support for the Record Validation filter, bringing additional schema formats into proxy-side enforcement.
ImplementationQuarkus LangChain4j · Merged
Authored a publisher fix and regression test so changed Agent Cards are republished even when the declared agent version stays the same.
ReliabilityApicurio Registry · Public record
Participated in the Sandbox application, including the earlier deferral and the renewed application that led to acceptance.
StewardshipProjects & implementations
Core maintainer
An open-source home for schemas and API definitions, with versioning, validation and compatibility rules for independently evolving applications.
Runnable integration
A connected MCP-to-A2A workflow that separates endpoint discovery, application orchestration and Agent Card publication checks.
Integration tests
A released Kroxylicious proxy, a schema registry and assertions for invalid records, mixed batches, bypass routes and dependency failures.
Executable example
A real Registry compatibility check paired with a feature-normalization test: an unchanged numeric type can still carry a different meaning.
Kubernetes lab
Follow a database change through Debezium, Kafka, an Iceberg commit and a Trino query. Diagnose data progress independently of process health.
Protocol comparison
The same deterministic capability behind two official SDKs, comparing output, errors, progress and caller-visible cancellation.
Security integration
Keycloak, Strimzi OAuth and Registry permissions exercised through positive controls, denied writes and broker-content checks.
Release lab
Version selection, cache refresh and rollback against Registry, with recorded local-model evaluations and explicit promotion criteria.
Measured experiment
An isolated Spark Structured Streaming experiment examining how processing delay and source admission affect prediction freshness.
Recovery integration
A PostgreSQL-to-Elasticsearch trace through Debezium and Kafka, testing updates, deletes, replay and read-alias cutover.
Maintainer workshop
Public application records and a worksheet for connecting project claims to evidence, unresolved questions and owners.
Community playbook
An ownership-aware contribution workflow: distinguish waiting on authors from waiting on maintainers, and make the next action visible.