Apache Parquet powers much of the modern data ecosystem, but Java’s go-to library, parquet-java, brings heavy dependencies and a single-threaded reader. This talk introduces Hardwood, a new Parquet parser built from scratch in modern Java, exploring the challenges of parallel decoding, performance techniques like page-level parallelism and adaptive prefetching, profiling with JDK Flight Recorder, and practical lessons from using Claude Code for AI-assisted development. Whether you’re interested in file formats, Java performance, or the realities of building software with AI, there’s something here for you.
