shotamatsubara
Remote: Yes (Asynchronous / Written-first communication preferred)
Willing to relocate: No
Technologies: Linux, Docker, Python, Polars, DuckDB, SQLite (WAL), libjemalloc2, C/C++, CUDA, vLLM, Parquet, Bare-Metal HPC
Résumé/CV: https://github.com/Matsubara-CEO | https://shotamatsubara.com
Email: [email protected]
I am an Infrastructure & Data Architect specializing in zero-cloud-cost batch pipelines, air-gapped environments, and deterministic memory bounding for heavy enterprise workloads on bare metal.
Recent engineering benchmarks (HP Z4 G4: Xeon W-2155, 128GB ECC RAM, RTX 5060 Ti 16GB): - High-Throughput Batch Inference: Processed 10M records in 10.9h (~254.9 req/s) with 0.00% error rate under strict offline isolation (Docker --network none). - O(1) Memory Clamping: Clamped host RAM between 6.7GiB and 9.4GiB (zero swap) using LD_PRELOAD libjemalloc2 dirty page purging and Polars chunked streaming. - Crash-Resilient Storage: Engineered 3.4GB SQLite WAL two-phase commit pipelines verified across hard SIGKILL events.
Available for fixed-scope contract engagements, architecture consulting, or fractional infrastructure engineering. Happy to run a quick proof-of-value benchmark on a non-sensitive sample binary (LAS/LAZ, PCAP, MDF4, SEG-Y, BAM) to demonstrate pipeline performance.