Database Systems Evaluation Expert

What you'll do

Mercor is piloting a set of database engineering benchmarks for AI agents, and wants a handful of engine-level engineers to write them. Selected experts each build one self-contained task in their specialty: database-engine implementation, diagnosing a production problem, or migrating an application.

A finished task has five parts: a clear problem statement, a reproducible repository or environment, a working reference solution, tests, and grading criteria. The criteria are the hard part. The ad asks for checks that tell a genuinely robust solution apart from one that merely passes, covering correctness, failure handling, concurrency, retries and regressions, plus performance measures such as latency, throughput and memory use. You then iterate with technical reviewers until the task holds up.

If you have ever written a test that passed on a buggy implementation of MVCC or a replication protocol, you know why they want internals people for this.

Who fits

• 5+ years of hands-on database-engine internals work, or a PhD in database systems
• Deep experience in storage engines, query execution, optimisation, transactions, replication or related low-latency systems
• Strong implementation skills and the ability to explain correctness and performance trade-offs clearly
• Go is preferred

The ad is explicit about who does not qualify on that background alone: DBAs, SQL analysts, ETL engineers and application-only database developers. You do not need to cover every task area.

The application

A self-paced technical assessment of about 35 minutes, with written answers and a coding exercise; you control when to move between sections. You need a laptop or desktop with camera, microphone and full-screen sharing, your own editor and terminal, Python 3.8+, and a compiler for C, C++, Rust, Go or Java. A starter kit and public tests are provided. Outside AI help is not allowed during the assessment.

Timing matters more than usual: applications close Friday 16 October 2026 at 12:00 p.m. Pacific, and Mercor will take the first three qualifying applicants for the pilot. If it fills, completed assessments may be considered if the project grows.

What it pays

$2,200 per task (shown on the listing as $2.2K). The task is estimated at about 25 hours, which works out to roughly $88 an hour if the estimate holds, and is due by Friday 30 October 2026. Review and feedback follow submission. Payment schedule, method and eligibility terms are not published on this listing.

Worth knowing

Good:

• A fixed fee for one well-defined piece of work, with a clear deadline
• Real engineering, not grading: you design the benchmark
• No residence requirement is published
• A completed assessment may carry over if the project expands

Less good:

• Only three places in the pilot, filled first come, first qualified
• Tight window: apply by 16 October, deliver by 30 October
• Fixed fee means overruns beyond 25 hours are on you
• No payment terms, visa terms or follow-on work are published

About this listing

Posted by Mercor as a remote contract and read from the live listing on 12 October 2026, marked as a new opportunity. The fee, hours estimate, deadlines and requirements are the ad's own. See Mercor.