Software Engineer, C - Codebase Q&A

Weekday AI

1d•$130 - $130•Remote

About The Position

We are seeking experienced C engineers to build the dataset that teaches AI agents to reason about real codebases. Codebase Q&A is a reinforcement-learning environment for training AI agents to explore and reason about software repositories they have never seen before. You will build difficult codebase-exploration tasks: high-level engineering questions about foundational C systems that cannot be answered by reading a single file, and that require runtime evidence to resolve — the kind of question a senior engineer answers by actually investigating the system. This track works on some of the most consequential C codebases in production anywhere: PostgreSQL, Redis, git, OpenSSH, OpenSSL, curl, libuv, libgit2, libarchive, Open vSwitch, tcpdump, tmux and zstd.

Requirements

3+ years of professional software engineering experience, with substantial production C
Demonstrated ability to navigate a large, unfamiliar C codebase and explain how it behaves at runtime — not just what the source says
Comfort with the things that make C systems hard to read: preprocessor-heavy builds, platform conditionals, manual memory management, and autotools/CMake/Meson build graphs
Ability to use debuggers, sanitizers, tracing, and profilers to establish what the code actually does
Precise written English: the questions and rubrics you write are the product

Nice To Haves

Experience with databases, cryptography, networking protocols, or systems programming is highly relevant — the task mix is weighted toward architecture and system design (39%) and code onboarding (34%), with a meaningful security slice

Responsibilities

Select tasks from a pre-validated pool of C repositories, each pinned to a specific git commit and arriving with an engineering question, positive and negative rubrics, foils, and a golden solution
Explore the repository until you genuinely understand the subsystem in question, including the relevant PR or issue history
Rewrite the task's question so it defeats frontier coding agents while remaining well-defined and fairly answerable
Adjust rubrics and foils so they reward correct reasoning and reject plausible-but-wrong answers
Run the full validation loop in Studio — Check, Golden, Run, Analyze — and iterate until every stage passes
Document your work in the task README and submit for expert review, addressing reviewer feedback on returned tasks