Software engineer and researcher building Active Model — an autonomous research system for optimization, algorithmic search, and systems-performance problems.
Active Model starts from a task, evaluator, and verification criteria, then investigates the problem, forms and tests hypotheses, implements candidates, and verifies results.
Recent work applies Active Model to mature, heavily optimized production software.
-
llama.cpp — PR #27478 Reworked batch-1 CPU flash-attention decode and large-allocation handling. On Qwen3-30B-A3B Q4_K_M at 8192-token context, end-to-end token-generation throughput improves +15.29% on AMD Zen 5 and +9.22% on ARM Neoverse-N1. The attention optimization alone contributes +10.67% and +7.43%.
-
zstd — three optimization PRs submitted upstream #4729 · #4732 · #4733 Includes a decompression-hot-loop optimization removing a loop-carried memory dependency, improving decode throughput by +2.7–3.4% with GCC and +5.5–9.3% with Clang on Zen 5, with the same direction on Intel Raptor Lake.
-
dav1d — two AV1 decoder optimizations prepared for upstream submission One removes an unnecessary compound-prediction scratch-buffer round trip, reaching +2.79% whole-decoder throughput on x86 and +1.73% on ARM Neoverse-N1. A second exact transformation removes 77.5% of redundant temporal-MV candidate processing. Both pass 240/240 AOMedia AV1 conformance vectors.
-
Solved the English perfect 10-square problem, open for more than 120 years. Code and result · History
-
Improved on Google DeepMind AlphaEvolve results on two packing problems, verified using AlphaEvolve’s published verifier:
- 32 variable-radius circles in a unit square
- 21 variable-radius circles in a perimeter-4 rectangle
-
Produced additional best-known results in circle packing and spherical codes, and a large-scale AC optimal-power-flow result accepted into MINLPLib.
Active Model explores whether frontier models can operate as autonomous computational researchers rather than only as coding assistants.
Its core principle is simple:
Use externally checkable objectives, explicit verification, and repeated hypothesis → implementation → measurement loops to search for real improvements.
Current work spans mathematical optimization, algorithm discovery, and performance engineering in production C/C++ software.
AI-assisted systems optimization · inference performance · autonomous research systems · program analysis · combinatorial and numerical optimization · scientific discovery
I am interested in research-engineering roles, collaborations, and technology-transfer opportunities involving autonomous optimization, high-performance software, and AI-for-systems.