This document summarizes a talk on using Apache Kafka and Avro for stream processing at Mate1. It discusses: 1. How Mate1 was using message queues before to address latency and scalability issues, but wanted improvements in type safety, versioning, and being rewindable. 2. How they used Apache Avro for serialization to gain type safety when producing and consuming from Kafka topics. 3. How they developed a simple data format and integrated an Avro schema repository to allow for versioned schemas and backwards compatibility when schemas change. 4. How the integration allows for rewinding by mapping offsets to points in time, and rebuilding state after crashes by reprocessing data from a previous