Accounting Expert: AI Model Evaluation

Huzzle Labs partners with frontier AI labs to evaluate and improve frontier models through deep human expertise. Contributors directly assess advanced AI systems in their specialist fields, receive competitive compensation, and help define the standards used to measure model performance.

Huzzle Labs is seeking experienced accountants to write realistic accounting tasks that test how well frontier AI systems handle accounting work. You will create prompts that are difficult enough for a frontier model to fail, then write a scoring rubric for each task so reviewers can assess the output consistently. A rubric is a list of objective, checkable criteria that a reviewer marks pass or fail. Your judgement helps define the standards used to measure model performance.

This is an expert judgement role focused on realistic task design, careful review of model output and precise, consistent documentation. Training is provided. Prior AI training or data-annotation experience is required.

• Write realistic accounting tasks: Create requests of the kind a client, controller or business owner would send, covering month-end close, bank and credit card reconciliations, AR/AP ageing, accruals and adjusting entries, sales tax, job costing and the clean-up of incomplete or inconsistent books.
• Build from real material: Review a deal room containing hundreds of files, select the relevant documents and use them to develop prompts and rubrics for tasks a frontier model struggles to complete correctly.
• Require usable deliverables: Each request should call for a working output, usually a spreadsheet, a journal entry schedule or a short memo.
• Test against a frontier model: Run each request and review the output with the care you would apply when signing off accounting work. Where the model succeeds, increase the difficulty until a genuine gap appears.
• Write scoring rubrics: Define objective criteria that another reviewer can apply without repeating your work, for example, “the reconciled ending balance equals $X” and “the $X deposit is recorded as a customer deposit liability, not revenue.”
• Apply guidelines consistently: Follow the task and rubric guidelines, flag ambiguous cases and help refine the criteria.

• At least 5 years in accounting, bookkeeping or finance operations.
• Prior AI training or data-annotation experience is required.
• Hands-on knowledge of month-end close, reconciliations, AR/AP, accruals and adjusting entries, sales tax and job costing.
• Excellent written English and precision with numbers.
• Careful and detail-oriented: accurate figures and objective, checkable criteria.
• Available 20 to 40 hours per week for an initial one-week trial, with potential extension.

For a client deal room covering a small business with incomplete and inconsistent books, you would:

1. Review the files and select the relevant documents, such as bank statements, ledgers, invoices and ageing reports.
2. Write a prompt as the business owner or controller would, asking for a reconciliation and the adjusting entries needed to close the month.
3. Run the prompt against a frontier AI model and check the spreadsheet or journal entry schedule line by line.
4. If the model completes the task correctly, increase the difficulty, for example, by adding a customer deposit, a timing difference or a sales tax complication, until it fails.
5. Write the rubric, such as “the reconciled ending balance equals $X” and “the $X deposit is recorded as a customer deposit liability, not revenue,” so another reviewer can grade any answer without repeating the work.

An NDA must be signed before the engagement begins.

Compensation: $60 - $90 USD / hour

Find more English Speaking Jobs in Germany on Arbeitnow