Compiler Developer & OS Architect
Building compilers, operating systems, and GPU engines from scratch — no LLVM, no GCC, no frameworks
Lima, Perú 🇵🇪 | 6+ years Systems R&D | Binary Is Binary 💀🦈
"Binary Is Binary — no runtime, no intermediaries, no excuses"
| Project | Language | Output | Key Achievement |
|---|---|---|---|
| ADead-BIB v8.0 | C / C++ | x86-64 native | 354,134 lines proven IR |
| PyDead-BIB v4.0 | Python | x86-64 native | 0.305ms · no CPython · no GIL |
| FastOS v4.0 | C / ASM | Bare metal | 89KB kernel · boots in QEMU |
| Metal-Dead | Python | Native AI | 0.60MB AI · no PyTorch |
Python → x86-64 native. No CPython. No LLVM. No GCC. No runtime. No GIL.
pyb run archivo.py # compile → RAM → execute → 0.305ms
| CPython 3.13 | PyPy 7.3 | Nuitka | PyDead-BIB | |
|---|---|---|---|---|
| Runtime | ❌ 30MB | ❌ 200MB | ❌ 8MB | ✅ 0 bytes |
| GIL | ❌ | ❌ | ❌ | ✅ eliminated |
| No LLVM/GCC | ✅ | ✅ | ❌ | ✅ |
| AVX2 SIMD | ❌ | ❌ | ❌ | ✅ auto |
| UB compile-time | ❌ | ❌ | ❌ | ✅ 13+ types |
| Hello World | 30MB | 200MB | 8MB | ~2KB |
| Startup | ~50ms | ~2000ms | ~10ms | 0.305ms |
JIT Killer v2.0:
"The CPU doesn't think — it already knows.
The RAM doesn't wait — it already receives."
Pre-resolved dispatch table → instant image → VirtualAlloc → JMP
CPU features: AVX2 ✓ SSE4.2 ✓ BMI2 ✓
GPU: RTX 3060 → 24,119 GFLOPS from compiled Python
86/86 tests PASS
→ github.com/AndreeSalazar/PyDead-BIB
FastOS v4.0 — QEMU boot:
CPU: AuthenticAMD ✅
BG: KERNEL Ring0 APPROVE ✅
AVX2 SSE4.2 AES ✅
PIC OK PIT 100hz ✅
MBR: 512 bytes (FASM)
Kernel: 65,536 bytes (C99) ← 64KB kernel
Image: 10,485,760 bytes
16→256 bit gradual CPU wake — no Linux, no Windows — bare metal.
|
|
|
|
struct EddI {
title: "Compiler Developer & OS Architect",
years: "6+ years (2019-2026)",
location: "Lima, Perú 🇵🇪",
compilers_built: vec![
"ADead-BIB v8.0 — C/C++ → x86-64 (354K lines)",
"PyDead-BIB v4.0 — Python → x86-64 (0.305ms)",
"ADead-GPU — GPU deterministic compiler",
"Zom — ADead-BIB + ROCm/HIP",
],
os_built: vec![
"FastOS v4.0 — complete OS, boots in QEMU",
"Custom bootloader (FASM, 512 bytes MBR)",
"64KB C99 kernel — memory, scheduler, IRQ",
],
gpu_achievements: vec![
"1,000,000 entities @ 75 FPS (RTX 3060)",
"24,119 GFLOPS from compiled Python",
"REACTOR: Vulkan 98% less boilerplate",
"EPICX: DirectX 12 framework",
],
philosophy: "Binary Is Binary — no intermediaries",
}Rust (Advanced) • C / C++ • Assembly x86-64
Lexer / Parser / AST / IR / Codegen
PE / ELF binary generation — no linker
ISA x86-64 encoding • FASM • NASM
Bootloaders (FASM) • Kernels (C99)
Memory Management • Scheduling • Interrupts
x86-64 Long Mode • CPUID • APIC • PIT
Bare metal — no Linux — no Windows dependency
Vulkan API • DirectX 12 • CUDA • wgpu
Compute Shaders • SPIR-V • HLSL • GLSL
GPU-Driven Architecture • Real-time Simulation
SIMD (AVX2 / SSE4.2 / BMI2)
JIT Killer v2.0 — VirtualAlloc executor
UB Detection — compile time
Deterministic execution • Cache optimization
PyDead-BIB v4.0 — Python Native Compiler 🆕
github.com/AndreeSalazar/PyDead-BIB
First Python compiler generating native x86-64 without CPython, LLVM, GCC, or any runtime.
- 0.305ms startup vs CPython's 50ms
- 2KB Hello World vs 30MB CPython runtime
- JIT Killer v2.0: pre-resolved dispatch, instant image, VirtualAlloc executor
- 13+ UB types detected at compile time
- AVX2 SIMD auto-vectorization
- CUDA + Vulkan GPU dispatch from Python
- 24,119 GFLOPS RTX 3060 from compiled Python
- 86/86 tests PASS
- Techne License v1.0 — free personal/OSS, 10% royalty commercial
"The CPU doesn't think — it already knows.
The RAM doesn't wait — it already receives."
ADead-BIB v8.0 — C/C++ Native Compiler
github.com/AndreeSalazar/ADead-BIB
Complete C/C++ compiler generating machine code without GCC, LLVM, or external linkers.
- 354,134 lines of proven compiler infrastructure
- IR ADeadOp SSA — own design
- PE (Windows) + ELF (Linux) + FastOS .po output
- Full OOP: classes, inheritance, polymorphism, vtables
- UB detector: 21+ types
- 16 → 256 bit targets (SSE → AVX2)
- Binary Guardian stamp
FastOS v4.0 — Complete OS from Scratch
OS built from zero — boots in QEMU.
- Custom bootloader (FASM, 512 bytes MBR)
- 64KB C99 kernel
- Memory management, scheduler, interrupts
- AVX2 SSE4.2 AES detection at boot
- 16→256 bit gradual CPU wake
- Target: bare metal without Linux or Windows
GPU-Driven Massive Simulation — 1M Entities @ 75 FPS
github.com/AndreeSalazar/GPU-Driven
| Entities | Sim Time | FPS |
|---|---|---|
| 100,000 | ~0.05ms | 120+ |
| 500,000 | ~0.10ms | 100 |
| 1,000,000 | ~0.14ms | 75 |
Deterministic: same input → same output guaranteed.
REACTOR — Vulkan Framework (98% less boilerplate)
github.com/AndreeSalazar/REACTOR-Framework-for-Vulkan-
- Visual node editor (Blender-style)
- 150-330KB executables
- 0.36s load time
- Zero heap allocations at startup
- Hot-reload SPIR-V shaders
Self-Directed Systems R&D (2019-2026)
- 6+ years intensive compiler + OS + GPU development
- 50+ public repositories
- Production-quality deliverables
Universidad Nacional Federico Villarreal
- Computer Engineering (In Progress)
- Admission: Special modality (Law 29973)
Everyone uses LLVM. I don't.
Everyone uses GCC. I don't.
Everyone uses runtimes. I don't.
Everyone keeps the GIL. I removed it.
Everyone uses JIT to learn. Mine already knows.
354,134 lines of proven compiler infrastructure.
89KB OS kernel that boots.
Python executing in 0.305ms.
1M entities at 75 FPS.
1 developer. Lima, Perú. 🇵🇪
Binary Is Binary. 💀🦈
💼 Open to: Compiler Engineering · OS Development · GPU Computing · Performance Engineering
ADead-BIB 💀 C/C++ → PyDead-BIB 💀 Python → FastOS 💀 Bare Metal
IR ADeadOp SSA
Binary Is Binary
Eddi Andreé Salazar Matos
Lima, Perú 🇵🇪