Importing documents: upload, extraction, review and save
How import works
- Upload a file, or paste text for the starter import.
- Extraction reads the file and proposes a draft record. The draft is not saved.
- Review every proposed field against the original. Correct what is wrong and fill in what is missing.
- Save. Only then is a record created.
Extraction proposes values from the text of the document. It does not check that a document is genuine, valid, current or sufficient. You decide that. Soteriaa does not give legal or technical advice. See What Soteriaa does and does not do.
Extraction uses an AI model. A document that is blank, scanned without readable text or hard to read may produce an incomplete proposal, or none.
Import routes by workspace
| Workspace and route | Accepts | What is proposed | What happens to the file |
|---|---|---|---|
| Welding: qualification certificate import | PDF only. Up to 20 MB. | A welder and a qualification (see the field list below). WPS and WPQR extraction is not available yet. | The PDF is uploaded and stored when you run extraction. The welder and qualification records are created only when you confirm. |
| Welding and sponsor: starter import (pasted text) | Pasted text, up to 20,000 characters. | Welding: welders, qualifications and continuity confirmations. Sponsor: workers, right-to-work checks, evidence, absences and change events. | No file is involved. On save, only welders (welding) and sponsored workers (sponsor) are created. Other reviewed items are listed as skipped so you can add them by hand. Duplicates are detected and not created again. |
| Welding: document library upload | PDF or .docx. Up to 10 MB. Old .doc files are not supported. | Text is extracted for search and an AI suggestion of the document type, a short summary and a likely matching record. | The file is stored and listed as soon as it is uploaded, with the suggestions attached. No welder or qualification record is created. |
| Sponsor: evidence upload with extraction | .pdf, .png, .jpg, .jpeg, .webp. Up to 20 MB. | For six evidence categories, field values plus signals on whether the document type and the name match the worker (see below). | The file is stored as evidence on the worker. Proposed values are applied only if you confirm them. |
| RAMS, HR and commercial: import a record | PDF or .docx. Up to 10 MB. 100 imports a month on every plan. | One record: the record type, a title, dates and the data fields for that type (see below). | The import step does not store the file. Attach it to the record as evidence if you want to keep it. |
Welding: fields proposed from a qualification certificate
- Welder: full_name, identification, date_of_birth.
- Qualification: certificate_number, standard, welding_process, material_group, product_type, joint_type, position, thickness_min, thickness_max, diameter_min, diameter_max, backing, date_qualified, examining_body, revalidation_basis, revalidation_due, next_continuity_due, notes.
- Continuity: confirmed_on, confirmed_by, method, next_confirmation_due.
- Confidence: an overall score and short notes on uncertain or missing values.
A save is blocked until the draft has a welder name or identification, a certificate number, a qualification standard and at least one other qualification detail such as the welding process. That check only confirms the fields are filled in. It does not check the certificate. A document that looks like a blank template is flagged.
Ranges are entered as written on the certificate. Soteriaa does not calculate or approve qualification ranges by code.
Sponsor: fields proposed from evidence
| Evidence category | Fields proposed |
|---|---|
| Passport / identity | full_name, document_number, expiry_date, nationality, date_of_birth |
| Visa / BRP / eVisa evidence | full_name, visa_type, document_number, expiry_date, start_date |
| CoS assignment confirmation | full_name, cos_number, visa_type, start_date |
| Right-to-work evidence | full_name, share_code, check_date, work_permission_expiry |
| Share code result | full_name, share_code, check_date, work_permission_expiry |
| Salary evidence | full_name, employer, gross_pay, net_pay, pay_period, annual_salary |
Other evidence categories are stored without structured fields. For every category the model is asked only whether the document is the expected kind and whether the name matches the worker. It does not judge validity or sufficiency.
RAMS, HR and commercial: fields proposed
| Workspace | Record type | Data fields |
|---|---|---|
| Health and safety / RAMS | Site / project | Location, Client, Project scope |
| Health and safety / RAMS | Assessment / activity | Activity, Location, Hazards supplied, Controls supplied, Assessor, Evidence references |
| HR and employment | Employee / candidate | Job title, Department, Employment status, Manager, Work location, Probation date, Appraisal date |
| HR and employment | Employment document / review | Document kind, Jurisdiction supplied, Objectives supplied, Actions supplied, Probation date, Appraisal date |
| Commercial forms | Customer / supplier | Party type, Address, Contact reference |
| Commercial forms | Commercial document | Document kind, Project reference, Currency, Items, Tax rate, Payment reference, Amount supplied |
The proposal also covers the start, end, review and due dates for the record type. It never chooses a status, because a status is a fact you record: a document is only paid or approved because you say so. It never links a parent record. You choose both.
What you must check
- Compare every proposed value with the original document, especially names, numbers and dates.
- Fill in anything the proposal left blank.
- Decide yourself whether the document is genuine, current and fit for its purpose. Soteriaa does not.
- Keep the original with your own controlled records. Follow your own document-control and retention procedure.
Frequently asked questions
Is anything saved automatically?
No record is created from an import until you review and confirm it. Two things are stored earlier: the welding qualification PDF is uploaded when extraction runs, and a file uploaded to the welding document library is stored and listed immediately.
Can I import a scanned PDF?
A scanned PDF with no selectable text is stored in the welding document library but cannot be searched, because OCR is not implemented. The record imports read the document with an AI model and may or may not be able to read a scan, so check the result.
What if the extraction is wrong?
Correct it in the review step before saving, or enter the record by hand. If extraction is unavailable, the screens say so and you can enter the record yourself.
Try it on your own records.
Your first records and due dates take a few minutes to set up.