~/gandhiji — available for work
I build things close to the metal and close to the math — from systems software in C, C++ and Rust to numerical work in Julia and R, all driven from the Unix shell.
01 — stack
Ten languages, one mental model: understand the machine, then write the smallest thing that explains it.
Zero-cost abstractions, explicit memory, predictable latency. I care about what the compiler actually emits.
pub fn parse(buf: &[u8]) -> Result<Frame, Err> {
let (hdr, rest) = Header::read(buf)?;
// no copies, no surprises
Frame::decode(hdr, rest)
}Simulation, statistics and plots that hold up to review.
Scripts that fail loudly and run quietly.
#!/usr/bin/env fish
for f in (fd -e rs)
cargo fmt -- $f
endClean, semantic markup — fast by default, readable by everyone.
02 — work
Systems, tooling and simulation — a few pieces that shipped and are still running.
A cache-friendly N-body integrator in Rust with SIMD inner loops and a Julia reference implementation for validation.
An eBPF-backed profiler that traces scheduler latency on Linux and renders flame graphs without a runtime dependency.
Reproducible coastal-surge modelling: R for the statistics, C++ for the solver, one container to run the whole pipeline.
03 — approach
Profilers, traces and benchmarks first. Intuition is a hypothesis, not a result.
The right language for the job — not the newest one. Often that's a shell script.
Pinned versions, one command to run, output you can diff. Future-you is a user too.
04 — contact
Open to systems, scientific and tooling work. Replies within a day.
hello@gandhiji.dev