JG.
Jaret Eduardo González Carrasco

High-Performance
Computing & AI

Building efficient intelligent systems through parallel computing, GPU acceleration, and systems programming.

C++RustCUDAPython
View ProjectsRésuméGitHubLinkedIn
Core Specializations

Where hardware and software meet.

01

High-Performance Computing

Parallel workloads, profiling, and optimization for computation that has to scale.

  • Parallel workloads
  • Profiling
  • Optimization
  • Scalable computation
02

AI Systems

Performance-aware infrastructure for AI workloads, from data movement to inference throughput.

  • Performance-aware AI infrastructure
  • Efficient computation
  • ML systems
03

GPU Computing

Heterogeneous CPU/GPU acceleration built on CUDA and parallel execution models.

  • CUDA
  • Heterogeneous computing
  • CPU/GPU acceleration
  • Parallel execution
04

Systems Programming

Memory, concurrency, and compilers — the layer where correctness and performance meet.

  • Memory management
  • Concurrency
  • Linux
  • Compilers
Technical Stack
CC++RustPythonCUDA
Systems Tooling
LinuxCMakeGCCLLVMGDBValgrindGNU MakeBash
Concepts
Parallel ComputingMultithreadingConcurrencyGPU AccelerationMemory ProfilingCache OptimizationPerformance EngineeringPOSIXCross-compilation
Featured Projects

Systems built for parallel, high-throughput computation.

Flagship Project

ParallelPilot

AI-powered automatic parallelization for serial Python workloads.

Problem

Serial Python simulation workloads leave available cores idle, and manually parallelizing them is time-consuming and error-prone.

Approach

ParallelPilot analyzes serial Python code, uses an LLM-driven pipeline (Codex/GPT) to propose a parallelization strategy, and rewrites the workload to execute across cores with multiprocessing and joblib — then numerically verifies the parallel output against the original serial result before trusting it.

Architecture
  • Static analysis of the serial workload
  • LLM-driven parallelization strategy
  • Codegen with multiprocessing / joblib
  • Numerical verification against serial baseline
  • Benchmark-driven acceptance
PythonmultiprocessingjoblibCodexGPT
Serial
Analysis
LLM-driven parallelization strategy
Parallel
Core 0
Core 1
Core 2
Core 3
Numerical Verification
Parallel output checked against serial baseline
Up to 8×

on tested workloads

KAIML

High-performance ML computing library.

Approach

KAIML is a C++20 and Rust computing core with CUDA acceleration, dispatching work across CPU threads and GPU blocks for heterogeneous execution.

Architecture
  • CPU workload intake
  • Dispatch across CPU threads and GPU blocks
  • Heterogeneous parallel execution
  • Result aggregation
C++20RustCUDAMultithreadingGPU Acceleration
CPU Workload
Dispatch
Split across CPU threads and GPU blocks
CPU Threads
GPU Blocks
~3×

ML execution improvement, tested workloads

KAII-CC

Custom compiler architecture.

Approach

KAII-CC is a C++17 compiler built end-to-end: lexical analysis, parsing, AST construction, and native code generation, with manual memory management profiled and validated with Valgrind.

Architecture
  • Lexical analysis
  • Parsing
  • AST construction
  • Native code generation
  • Manual memory management
C++17Valgrind
SourceLexerTokensParserASTCodegenNative
~30%

optimization improvement

KAI

High-performance AutoML computational core.

Approach

KAI is a C++ computational core built around concurrency, cache-aware memory layout, and high-throughput processing for large-scale AutoML workloads.

Architecture
  • Large data block intake
  • Cache / memory-aware layout
  • Concurrent execution lanes
  • High-throughput processed output
C++ConcurrencyMemory Efficiency
Large-Scale Data
Cache-Aware Memory Layout
Concurrent Execution Lanes
50 GB/day

sustained processing throughput

Hackathons
HackMTY 2024

Caudal IA

Predictive maintenance system for industrial machinery using AI to identify anomalies and potential equipment failures.

HackMTY 2025

Nibly

Operational intelligence and predictive insights for airline catering operations.

Experience

Engineering roles and impact.

Software Engineer

API Inmobiliaria

December 2025 — Present

  • Optimized data pipelines and refactored legacy code in Linux environments.
  • Applied performance-oriented data structures and low-level optimization techniques.

~20% — system throughput improvement

About

I care about how software actually uses the hardware underneath it, and how much faster and more efficient computation can become when we use that hardware well.

My work sits at the intersection of high-performance computing and AI: parallelism, GPU acceleration, memory behavior, and the systems-level decisions that determine whether a workload takes minutes or seconds.

Education

  • B.S. Software Engineering

    Universidad Tecnológica de Torreón

    Expected December 2027

  • TSU Software Development

    Universidad Tecnológica de Torreón

    April 2026

Contact

Let's build something fast.

Open to conversations about high-performance computing, GPU acceleration, and systems engineering roles.

EmailLinkedInGitHubRésumé

Jaret Eduardo González Carrasco — High-Performance Computing & AI

Built with Next.js & WebGL