rebrew
Compiler-in-the-loop decompilation workbench for binary-matching game reversing.
- 16
- Python
- MIT
- 2 h ago
maci0
Container runtimes, Kubernetes internals, LLM inference down to the metal, and reversing old game binaries. Senior Principal Architect @RedHat.
Built from maci0's public GitHub profile, Singapore. Not on Take My Repo yet, so there are no posts or updates here.
Compiler-in-the-loop decompilation workbench for binary-matching game reversing.
An open source reimplementation of the Miles Sound System for older games
Unofficial Home Assistant custom integration for SP Group electricity and water
Thermal mitigation for NVIDIA GB10 / DGX Spark / ASUS Ascent GX10: temperature-driven GPU clock governor, thermal logger, clock cap. Works around the under-load power-off issue.
Standalone docker toolchain images for legacy Windows/DOS compilers (MSVC 1.0-11.0, Borland, Watcom, Delphi) — build sources, pinned manifests, wrappers. No compiler binaries in this repo.
Self-improving AI agent harness in Zig 0.16 — gated self-patch loop, sandboxed WASM tools, CLI + web UI
Virtual baremetal OpenShift lab on libvirt: sushy Redfish BMC emulation, agent & IPI installs, bonded NIC + VLAN topology, two VIPs, split-DNS, and in-cluster BareMetalHost power management, all from one data-driven Ansible topology.
Pure-C Qwen3.8-27B CPU inference: mmap Q4 GGUF, AVX-512 kernels, GatedDeltaNet hybrid, lossless DFlash2 + n-gram speculative decoding
Qwen3.8-Flash-Next on 2x DGX Spark: research, llama.cpp GGUF serving configs, measured results (GPU experts + PLE streamed from disk, 1M ctx, MTP analysis)
A memory-efficient Zig reimplementation of Moonraker, the API server for Klipper.
Byte-identical C89 reconstruction of Europa 1400: The Guild (server.dll), matched with rebrew and MSVC 6.
Caveman mode for DeepSeek Harness: shorter prose, same meaning, with code, commands, paths, and error strings left intact.
Pure-Python drop-in subset of Ray's API that makes vLLM distributed inference work across nodes. ~1,400 LoC, one dependency, bind-mount into the stock vLLM image. NVIDIA + AMD, validated cross-node.