Unsafe, FFI & the Edges

required1,291 repos of evidence

Systems roles assume the same base layer as every other Rust job — plus comfort at the edges where safe Rust meets the operating system.

The destination — non-negotiable
  • understand what unsafe blocks promise and what they cost
  • cross an FFI boundary in both directions without UB
  • review unsafe code critically instead of trusting it
+ beyond Cargo: ownership deeply · unsafe fundamentals · FFI basics · no-panic discipline
The vehicles — 2 ways in, any combination works

Crates that wrap C libraries safely — the canonical unsafe-review reading list.

Prove it — build this

Write safe Rust bindings over a small C library — document the invariants each unsafe block relies on.

Can you honestly tick these?

Memory & Data Layout

required1,304 repos of evidence

Zero-copy buffer handling is the recurring pattern across the corpus's runtimes, proxies, and databases — the skill that makes Rust worth hiring for.

The destination — non-negotiable
  • understand stack vs heap, cache lines, and layout costs
  • process data zero-copy where it matters and prove the win
  • profile allocations instead of guessing
+ beyond Cargo: stack vs heap · cache lines & locality · zero-copy parsing · allocation profiling
The vehicles — 3 ways in, any combination works

Repos built on bytes/memmap2/bytemuck — where buffer discipline is the architecture.

Prove it — build this

Build a binary file parser (e.g. a WAL or image format) that's zero-copy where possible — benchmark it against a naive version.

Can you honestly tick these?

Concurrency & Parallelism

required952 repos of evidence

These crates are the corpus's standard answers to shared state — knowing which one a problem needs is the systems interview in miniature.

The destination — non-negotiable
  • understand what Send and Sync actually promise the compiler
  • choose between channels, locks, and lock-free structures and justify it
  • measure parallel speedups instead of assuming them
+ beyond Cargo: Send/Sync semantics · lock-free vs locked · work stealing · atomics & ordering (basics)
The vehicles — 2 ways in, any combination works

rayon-powered tools — embarrassingly parallel work done properly, with measurable speedups.

Prove it — build this

Build a parallel grep or log analyzer: rayon for scanning, crossbeam channels for results, a benchmark table in the README.

Can you honestly tick these?

Networking & Protocols

required259 repos of evidence

The corpus's proxies, VPNs, and RPC layers all live here — TLS with rustls and hand-rolled framing over tokio are their common denominator.

The destination — non-negotiable
  • understand what happens between connect() and the first byte
  • implement a wire protocol with framing you can defend
  • run TLS on both ends and know what the handshake costs
+ beyond Cargo: TCP/UDP semantics · TLS handshakes · protocol framing · epoll/io_uring awareness
The vehicles — 3 ways in, any combination works

QUIC, TLS, and raw-socket codebases — the transport layer read from source.

Prove it — build this

Write a TCP chat or key-value server with a documented wire protocol, TLS, and a stress-test client — numbers in the README.

Can you honestly tick these?

Performance Engineering

recommended453 repos of evidence

Measured claims beat adjectives. Corpus projects that win adoption publish criterion benchmarks — do the same in your portfolio.

The destination — non-negotiable
  • benchmark honestly and know why microbenchmarks lie
  • read a flamegraph and act on it
  • make something measurably faster and document before/after
+ beyond Cargo: Linux perf tooling · flamegraphs · benchmark methodology · SIMD awareness
The vehicles — 2 ways in, any combination works

Repos with serious criterion suites — read how performance claims get earned.

Prove it — build this

Take any earlier project and make it 2× faster — publish the flamegraphs, the benchmark suite, and what didn't work.

Can you honestly tick these?

OS & Platform

differentiator1,565 repos of evidence

The differentiator: corpus tools that manage processes, namespaces, and files directly are where systems hires get made.

The destination — non-negotiable
  • understand containers in terms of namespaces and cgroups
  • know what fsync guarantees and what it costs
  • write software that survives kill -9 correctly
+ beyond Cargo: Linux syscalls · processes & signals · filesystems · containers from first principles
The vehicles — 3 ways in, any combination works

Supervisors, shells, and init systems — the process lifecycle owned directly.

Prove it — build this

Build a minimal process supervisor (start, restart, signal forwarding, log capture) using nix — no shelling out.

Can you honestly tick these?
At the end of this road

Open roles matching this destination right now:

Confirmed active on YC jobs board. Rust explicitly required. Compensation $50K–$80K + 0.10–0.30% equity. Full-time remote.

Open

Building a dynamic runtime on top of the Linux BPF sub-system. Looking for Rust developers with deep interest in Linux internals, distributed systems, and systems programming.

Open

Polars is hiring a performance-focused Rust engineer. Requirements include experience with a low-level language (preferably Rust), systems programming fundamentals, and profiling tools like samply.

Open
← All destinationsBrowse the evidence corpus →