Part Time
$500/mo
20
Oct 24, 2026
US lawfirm and media company that turns public-records responses into clean, searchable data seeks a REMOTE Research & Quality Director based in the Philippines.
An Example of the Legal Tasks
- Reviewing filings and file names
- Reviewing responses and software indicated values for lawsuit feasibility
An Example of the Data Cleaning task:
- Software reads thousands of pages of
THE JOB (you will be provided a Claude account)
- Check machine-extracted facts against the original pages and d
- Write down, in plain rules, how borderline cases get d
- D
- Measure accuracy by sampling, and report it honestly, including what you do not know yet.
- Train and supervise a small team of reviewers as we grow.
GOOD FIT
You have worked as with technical tools and data before. Checking a claim against its source is your daily habit. You do NOT need to code. You do need to understand, in general terms, how people check and improve automated systems that read documents: what a label is, why you test on examples the system has not seen, why one accuracy number can mislead.
WE LOOK FOR
- Strong written English (you will write a short reason for every decision)
- A habit of opening the source before accepting a claim
- Comfort with spreadsheets and counting things carefully
- Patience with messy documents: scans, forms, long policy manuals
- Good judgement about when to d
Nice to have: public-records or court documents, labelling/reviewing data for AI, legal or policy background.
THE POSITION
40 hours per week - London afternnon hours 1pm-9pm)
Remote
$500/mo
Availability on WhatsApp
Reports to the founder and the UK team. The documents contain confidential and personal information: you will keep everything confidential and work only inside our tools.
TO APPLY (we read your answers BEFORE your CV)
Send your CV and answer each question in ONE paragraph, in your own words, no more than 180 words each:
1) A system read 200 documents and extracted 2,000 facts. You check 50 of them against the pages and find 6 wrong. What do you conclude about the other 1,950? What would you do next, and what would you tell your manager?
2) You fix the system's mistakes on 30 documents. It now gets all 30 right, and a colleague says the problem is solved. What is wrong with that conclusion, and how would you check?
3) A system keeps marking a yes/no field "yes" whenever a document merely mentions the topic. You can only change the written instruction it is given. Write the new instruction in two or three sentences, and say how you would test that it works without breaking something else.
4) To save money, a program skips about 40% of the pages in very long manuals because they do not mention the topic. How would you find out whether it is skipping pages that matter, and what would you do with the answer?
5) Tell me why ClaudeCode is better than a traditional AI chat tool?
Shortlisted applicants do a short paid trial task of 2 to 3 hours and a five-minute follow-up call about their own answers. We hire on whether you ask a question when a case is ambiguous, not on speed.