Open-source data systems contributor working at the boundaries between query engines, lakehouse formats, and index infrastructure — turning correctness, durability, and performance problems into tested upstream changes in Rust and C++.
- Durability & correctness — WAL generations and fencing, object-store and cache lifetimes, weak-memory concurrency, safe FFI boundaries, NULL and overflow semantics.
- Index & query behavior — HNSW/IVF alignment against the literature, SQL and Parquet semantics, vectorized Arrow execution.
- Lakehouse interoperability — Iceberg, Paimon and DuckLake metadata and time travel.
- Durable state & distributed jobs — Doris × Lance index job lifecycle with fence, quota and edit-log replay, on JNI metadata reads with explicit native-allocator ownership.
- Cross-language boundaries — a panic firewall across 66 C entry points, and distributed segment build APIs that share one trained model across fragments.
- Engine semantics — Parquet
INTERVALBloom pruning, input immutability indecode, a mixed-engine watermark bug found and fixed upstream, and vectorized decimal execution (~9–11×). - Weak-memory concurrency — a publication race in Paimon C++ whose compiler-only barrier is silent on x86.
Selected upstream contributions: 54 merged pull requests across 14 repositories in the Apache, DuckDB and Lance ecosystems.