QA Engineer: Email Parsing API
Budget / SalaryHourly project
TypeFreelance project
LocationRemote
Posted11 hours ago
QA Engineer: API Test Harness for an Email-to-Data Platform
bind.co connects to a small business’s Gmail or Outlook inbox and turns what’s in it into structured records through a REST API: contacts and their roles, documents and their fields, expenses and more. I need an experienced QA engineer who will treat our API exactly as a third-party client would. You’ll build a lightweight test harness, seed test inboxes with realistic business mail, and run a scored test round on every release.
Milestone 1 (fixed price): build the harness
Create three test inboxes that you control: two Gmail and one Outlook. All mail, documents and contacts must be fabricated, with no real personal data.
Use the Gmail API and Microsoft Graph to seed each inbox repeatably with small-business traffic. That means invoices, bills, receipts, statements, credit notes, expense reports, customer inquiries, sales threads, contracts and forms, plus decoys bind.co should ignore, such as newsletters, logos and calendar invites.
Connect each inbox to bind.co through our normal OAuth consent flow, the same way a customer would.
Write the answer key before looking at bind.co’s output, so we know exactly what it should find.
Call our endpoints, compare the returned JSON against the answer key, and produce a scored report.
The primitives you’ll test:
Company profile: name, aliases, tax numbers, fiscal year end, currency, address
User profile: owner’s title, profession, short bio
Contacts: real counterparties vs noise, merged duplicates, name, email, company
Contact roles: client, vendor, accountant, lawyer, banker, team member and others
Document type: invoice, bill, receipt, statement, credit note, tax notice and others, or none
Document fields: bill-from, bill-to, number, dates, total, tax, currency
Expenses: one record per real expense, with vendor, date, total, currency and payment status
Drafts: suggested emails to a contact, where every fact must be true
The seeded data and scripted scenarios must cover these edge cases:
phone photos of receipts, link-only invoices, ZIP attachments, two currencies and a non-English document
re-sent and duplicate invoices, which must not create new records
a correction later in a thread, which must update the existing record
a burst of about 50 newsletters, which should create nothing
revoking consent, which must stop data from being served
human corrections made through the API, which must never be overwritten on the next run
Milestone 1 is done when:
one command seeds the inboxes (or confirms they’re seeded), runs every test and produces a report
the report scores each primitive on recall, field accuracy, invented items and duplicates, and lists each failing case with expected vs actual
the code is delivered in a repo we own, with a README that lets someone else run it
the first full test round has been completed and reported
Estimated effort is 50-70 hours.
After that (hourly): release testing
We release regularly, but only significant change will require testing and those may be once a month or every few weeks. Each of those releases needs a fresh run, a short internal report with scores per primitive and regressions since the last run, and any failing cases reproduced or scripted. Expect 10–15 hours per significant release.
I’m open to hiring two testers. If I do, one builds and owns the harness, and the second runs releases and adds test cases.
Tech stack
Flexible. Python, TypeScript, Postman collections plus code, or whatever you’re most effective in. The must-haves are repeatable seeding, clear pass/fail metrics and a one-command run.
API docs, sandbox keys, our full testing spec and the answer-key template are shared after the NDA is signed. You won’t get database access; everything goes through the public API.
How to apply
Start your proposal with your answers to these questions. Proposals that don’t will not be reviewed.
An invoice is re-sent a week later with a corrected total. What should your answer key say, and how would you score it?
How would you test that revoking OAuth consent stops data from being served?
Have you tested a system whose output is a prediction (AI, OCR, matching)? How did you score it?
Roughly how many hours would you estimate for Milestone 1, and what’s your hourly rate for release testing?
Please also link to a test suite, test plan or defect report you’ve written. Redacted is fine. Shortlisted candidates do a paid 2-hour trial: writing test cases for one primitive from a short spec excerpt.
Resume and references are also required.
NDA required to apply.
All test date, code and findings belong to bind.co
bind.co connects to a small business’s Gmail or Outlook inbox and turns what’s in it into structured records through a REST API: contacts and their roles, documents and their fields, expenses and more. I need an experienced QA engineer who will treat our API exactly as a third-party client would. You’ll build a lightweight test harness, seed test inboxes with realistic business mail, and run a scored test round on every release.
Milestone 1 (fixed price): build the harness
Create three test inboxes that you control: two Gmail and one Outlook. All mail, documents and contacts must be fabricated, with no real personal data.
Use the Gmail API and Microsoft Graph to seed each inbox repeatably with small-business traffic. That means invoices, bills, receipts, statements, credit notes, expense reports, customer inquiries, sales threads, contracts and forms, plus decoys bind.co should ignore, such as newsletters, logos and calendar invites.
Connect each inbox to bind.co through our normal OAuth consent flow, the same way a customer would.
Write the answer key before looking at bind.co’s output, so we know exactly what it should find.
Call our endpoints, compare the returned JSON against the answer key, and produce a scored report.
The primitives you’ll test:
Company profile: name, aliases, tax numbers, fiscal year end, currency, address
User profile: owner’s title, profession, short bio
Contacts: real counterparties vs noise, merged duplicates, name, email, company
Contact roles: client, vendor, accountant, lawyer, banker, team member and others
Document type: invoice, bill, receipt, statement, credit note, tax notice and others, or none
Document fields: bill-from, bill-to, number, dates, total, tax, currency
Expenses: one record per real expense, with vendor, date, total, currency and payment status
Drafts: suggested emails to a contact, where every fact must be true
The seeded data and scripted scenarios must cover these edge cases:
phone photos of receipts, link-only invoices, ZIP attachments, two currencies and a non-English document
re-sent and duplicate invoices, which must not create new records
a correction later in a thread, which must update the existing record
a burst of about 50 newsletters, which should create nothing
revoking consent, which must stop data from being served
human corrections made through the API, which must never be overwritten on the next run
Milestone 1 is done when:
one command seeds the inboxes (or confirms they’re seeded), runs every test and produces a report
the report scores each primitive on recall, field accuracy, invented items and duplicates, and lists each failing case with expected vs actual
the code is delivered in a repo we own, with a README that lets someone else run it
the first full test round has been completed and reported
Estimated effort is 50-70 hours.
After that (hourly): release testing
We release regularly, but only significant change will require testing and those may be once a month or every few weeks. Each of those releases needs a fresh run, a short internal report with scores per primitive and regressions since the last run, and any failing cases reproduced or scripted. Expect 10–15 hours per significant release.
I’m open to hiring two testers. If I do, one builds and owns the harness, and the second runs releases and adds test cases.
Tech stack
Flexible. Python, TypeScript, Postman collections plus code, or whatever you’re most effective in. The must-haves are repeatable seeding, clear pass/fail metrics and a one-command run.
API docs, sandbox keys, our full testing spec and the answer-key template are shared after the NDA is signed. You won’t get database access; everything goes through the public API.
How to apply
Start your proposal with your answers to these questions. Proposals that don’t will not be reviewed.
An invoice is re-sent a week later with a corrected total. What should your answer key say, and how would you score it?
How would you test that revoking OAuth consent stops data from being served?
Have you tested a system whose output is a prediction (AI, OCR, matching)? How did you score it?
Roughly how many hours would you estimate for Milestone 1, and what’s your hourly rate for release testing?
Please also link to a test suite, test plan or defect report you’ve written. Redacted is fine. Shortlisted candidates do a paid 2-hour trial: writing test cases for one primitive from a short spec excerpt.
Resume and references are also required.
NDA required to apply.
All test date, code and findings belong to bind.co
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.