namespace shlok {
constexpr auto role = "Founding AI Engineer @ Clera";
constexpr auto location = "San Francisco, CA";
std::vector<std::string> interests = {
"AI systems", "compilers", "databases", "inference",
"agents", "competitive programming", "quantitative systems",
"making fast things faster"
};
auto philosophy = R"(
understand the paper;
build the primitive;
measure the bottleneck;
remove the abstraction tax;
ship.
)";
}|
An N-dimensional Transformer built from scratch in C++17 after reading Attention Is All You Need—without Eigen, BLAS, or a deep-learning framework.
|
A modular machine-learning library written primarily in C++, with Python bindings and implementations spanning regression, classification, clustering, and neural networks.
|
|
A low-latency text cleaning, deduplication, chunking, and vectorization pipeline for production RAG workloads, combining Python ergonomics with C++ acceleration.
|
A Homebrew-installable CLI that runs code, catches terminal errors, and uses Gemini to explain what broke, why it broke, and how to fix it.
|
|
A growing collection of C++ solutions to Kattis problems—algorithmic puzzles, data structures, graph problems, dynamic programming, and the occasional fight with a time limit.
|
|
high-level idea
│
▼
research paper ──► reference implementation ──► profiler ──► optimized system
▲ │
└────────── benchmarks ◄──────┘
{
"role": "Founding AI Engineer @ Clera",
"location": "San Francisco",
"current_mode": "reading abstractions from the bottom up",
"usually_found": [
"implementing papers",
"profiling already-fast systems",
"solving competitive programming problems",
"asking what happens below the API"
],
"status": "still compiling"
}$ ./shlok --mode build
[OK] ideas loaded
[OK] profiler attached
[WARN] abstraction detected
[RUN] removing unnecessary layers...
exit status: still building