# AI & GPU Algorithms

A growing collection of **GPU/CUDA, machine-learning, AI, and numerical-computing**
projects — each self-contained, with a reproducible benchmark and a short
technical write-up. The goal is depth over breadth: implement a thing, measure
it honestly, and explain *why* the numbers look the way they do.

## Projects

### GPU / CUDA
- **[Parallel ODE Integrator](cuda-hpc/01_parallel_ode_integrator/)** — a batch of
  independent oscillators integrated on the GPU (one CUDA thread each) vs a
  single CPU core. Verified bit-for-bit correct, **up to ~960× speedup** on an
  RTX 2080 Super. Uses CUDA events to separate compute from PCIe transfer and
  compares Euler vs RK4. → [report](cuda-hpc/01_parallel_ode_integrator/REPORT.md)
  · [lessons](cuda-hpc/01_parallel_ode_integrator/LESSONS.md)

  ![CPU vs GPU benchmark](cuda-hpc/01_parallel_ode_integrator/benchmark.png)

### AI / ML
- `AI algorithms implementations/` — algorithm practice and HackerRank AI problems.
- *(more coming)*

## Layout

```
cuda-hpc/                     GPU/CUDA projects (CMake, one folder per project)
AI algorithms implementations/  AI/ML algorithm work
```

Each project folder has its own README with build/run steps and results.

## Tooling

- **Languages:** C++17 / CUDA, Python.
- **Build:** CMake; per-project `build.bat` / `run.bat` helpers on Windows.
- **CI:** GitHub Actions compiles the CUDA projects on every push.