Computer Architecture Researcher (Research Coding Tasks)

What you'll do

You start from an architecture paper whose code is public, and remove the implementation of the design it proposes. An AI agent is then asked to rebuild that design and beat the authors' results on the paper's benchmarks. You write the task statement, define held-out workloads and per-metric thresholds, and supply a hidden reference solution along with a second one that clears the bar. You review the agents' attempts afterwards.

The ad names caches, prefetchers, branch prediction and accelerators evaluated in simulators such as gem5 or ChampSim. That points to tasks whose score is something like IPC, miss rate or predictor accuracy over held-out traces, so you need to be able to set up a simulator environment others can rerun. The ad does not say which traces or simulator versions are provided.

What you need

• Published computer architecture research with released code; the ad names ISCA, MICRO, HPCA or ASPLOS, or comparable venues
• A PhD in computer science or a related field, completed or in progress
• A link to a paper you authored and its code, plus its headline result, both required on the application form

Nice to have

• First-author papers reporting measured improvements over a baseline
• Comfort with Docker and reproducible benchmark setups, which matters when simulator builds are fragile
• Depth in caches, prefetchers, branch prediction and accelerators evaluated in simulators such as gem5 or ChampSim

What it pays

The AfterQuery board headlines $80/hour for this role. The ad text itself gives a per-task figure instead: $200 per approved task, with each task taking a few hours. Payment only follows approval, so a task that is rejected earns nothing for the time spent. The listing does not publish payment frequency, minimum weekly hours or monitoring terms; the AfterQuery company page covers what is known about how contractors are paid.

Worth knowing

Good:

• You work from your own paper and code, so the domain knowledge is already in your head
• Fully remote and asynchronous, with no shifts
• Simulator-based papers map well onto a task with a single, automatic score

Less good:

• Pay is per approved task, so the hourly figure depends on how quickly you build and validate a task
• The bar for a task is high: you must supply a hidden reference solution and a second solution that beats the authors' results, which takes real engineering time
• The ad says candidates go through background checks, and the application needs a paper link, its code and a headline result up front
• The ad warns that candidates who do not meet the stated requirements are deprioritised, so apply only if you fit

About this listing

Read from the AfterQuery Experts board on 29 October 2026. This is one of twelve Research Coding Tasks roles that share a template across different research fields; this write-up covers the specifics of this field. The listing does not give an application deadline or a number of openings. See AfterQuery.