# Adam DePrince

> Adam DePrince is a Python and C++ systems programmer focused on finance,
> high-performance computing, and systems development. His open-source work
> emphasizes low-level performance, SIMD, market-data systems, and practical
> systems software.

Website: https://adamdeprince.com/
GitHub: https://github.com/adamdeprince
Last updated: 2026-07-22

## Primary Projects

- Goblin Core: https://goblin-core.dev
  C++23 Redis-shaped server built to reduce the memory cost of hot state and
  the transport cost of latency-sensitive state. RESP and opt-in typed SBE run
  across TLS-capable TCP, native XLIO Ultra, Unix-domain sockets, shared-memory
  rings, and polled RDMA. Version 0.10 adds broker-acknowledged Kafka writes
  with bounded backpressure, firehose replication, and whole-database replica
  bootstrap. A million-member set uses 36.2% less memory than the leanest
  compared server, synchronous shared-memory calls take about 220 ns, and
  native XLIO runs seven
  commands at 4.98-8.37 us median over 100 GbE. Across two old R820s on 40 Gb/s
  InfiniBand, its polled one-sided RDMA ring completes SBE and RESP2 PINGs at
  2.495 and 2.673 us p50, respectively.

- Goblin Store: https://goblin-store.dev
  NUMA-aware large-object cache, HTTP server, and streaming caching proxy. It
  speaks memcache over TCP or native InfiniBand/RoCE and tiers object heads,
  middles, and tails across RAM, SSD, and optional HDD. It matches memcached
  latency with 7.4x less RAM and delivers 118 QPS versus 70 for extstore on the
  disk-sized benchmark. Its native InfiniBand path reaches 38.2 Gbit/s.

- stride-align: https://github.com/adamdeprince/stride-align
  Smith-Waterman and Needleman-Wunsch alignment library. Reported speedups:
  1.75x faster than Parasail on AVX-512, 1.37x faster on AVX2, 7.5x faster by
  geometric mean on LoongArch LASX, and 22.4x faster on specific workloads.

- massive-speedup: https://github.com/adamdeprince/massive-speedup
  High-performance Massive.com flatfile parser, data store, and latency-aware
  market microstructure simulator. Flatfile iteration is reported as 5x faster
  than Python gzip+csv.

- rtta: https://rtta.dev
  Low-latency incremental technical-analysis toolkit designed for tick-forward
  data, with batch and real-time interfaces.

- negcycle: https://github.com/adamdeprince/negcycle
  SIMD-optimized Python library for finding profitable FX execution cycles.
  Identifies the most profitable cycle in 1.8 us on an 11th-generation Intel i7.

- fast-kalman: https://github.com/adamdeprince/fast-kalman
  C++ and Python Kalman filter library. Reported C++ speedups over OpenCV range
  from 7x to 235x, with Python-level speedups around 5x.

## Other Notable Work

- DragonArray: https://dragon-array.org
  Loongson optimization lab. Its NumPy patch set enables more 256-bit LASX
  paths with a reported 3.3x geometric-mean speedup, and it now also ships a
  BLAKE3 LoongArch SIMD port from the public loongarch-simd source branch.

- bsort: https://github.com/adamdeprince/bsort
  Fixed-record external sort library for large data files, available as a C++
  executable and Python library.

- mosh patches: https://github.com/adamdeprince/mosh
  Patches adding SSH-like forwarding, agent support, scp-like bulk copy, and
  trained-compressor support to mosh.

- CPython SIMD experiment:
  https://github.com/adamdeprince/cpython/tree/3.14-simd-experimental
  Experimental CPython branch with SIMD dispatch applied to codecs, string
  scanning, bytes handling, JSON, CSV, urllib parsing helpers, html.escape, and
  set-implementation experiments.

- NumPy LASX experiment:
  https://github.com/adamdeprince/numpy/tree/loongson-experimental
  Proof-of-concept NumPy port to Loongson LASX.

- goblin-cannon: https://github.com/adamdeprince/goblin-cannon
  Low-latency cross-Atlantic SDR market-data radio implementation with AES,
  Viterbi decoding, CRC checking, QCI encoding, oversample-based decoding, and
  AVX2/AVX-512 support.

## Toy Websites

- NAICS Code Finder: https://naics.goblinreactor.com/
  Semantic search for ranking six-digit 2022 NAICS industries from a
  plain-language description of a business activity, with batch lookup support.

- CPV Code Finder: https://cpv.goblinreactor.com/
  Semantic search for ranking 2008 Common Procurement Vocabulary codes from a
  plain-language purchasing requirement, with batch lookup support.

- Sloppy Resume: https://sloppyresume.com/
  A joke site with real math that compares two PDF resumes using
  Needleman-Wunsch alignment and scores how suspiciously alike they are.

- DragonArray: https://dragon-array.org/
  Loongson optimization lab publishing LoongArch64 wheels, source patches,
  benchmarks, and utilities, including NumPy LASX wheels and a BLAKE3
  LoongArch SIMD port.

- R820 Field Notes: https://r820.goblinreactor.com/
  TensorFlow compatibility builds and CPU guidance for pre-AVX2 x86-64
  machines, including separate AVX1 targets with and without F16C.

- Bluesky Pulse: https://bluesky.goblinreactor.com/
  Live mathematical sample of Bluesky's public feed showing word rates, trend
  forecasts, language activity, and recent posts in a C++23 and Wt dashboard.

## Contact

Email and phone are intentionally encoded on the HTML homepage. The public
GitHub profile is https://github.com/adamdeprince.

For more detail, download /llms-full.txt.