GuidesAPI EndpointsAPI Changelog
Live AI Demo
API Changelog
Improved

More accurate names, certifications and job sections in parsing

Résumé parsing, text parsing and Document OCR return more accurate names, certification names, contact details and job sections, for the ways of writing a résumé or a job posting listed below.

Action required

  • Now: none. POST /profile/parsing/file, POST /text/parsing and POST /text/ocr keep their parameters and their responses.
  • Later: none. Nothing in this update is deprecated.

What changed

  • Letter-spaced names (Quicksilver): a candidate's name set in letter-spaced capitals is read as words, so the first_name, last_name and full_name of its info are right far more often for such résumés; section headings set that way also read as words in the profile's text.
  • Certification names with a colon or a bracket: a name such as PMP (Project Management Professional) or Lean Six Sigma: Green Belt comes back as one certification instead of two pieces, in parsed résumés (the profile's certifications and those of its experiences and educations) and in POST /text/parsing.
  • Cleaner certification names (POST /text/parsing): no list separator at their end, their closing bracket kept, and line breaks or runs of spaces inside a name turned into one space, such as Führerschein (Klasse B) or Permis B.
  • Entity edges (POST /text/parsing): entities come back without a list separator (, or ;) or a line break at their start or end, with both parsing models, and their start and end in entities leave it out; skill and certification names that end in a hyphenated or multi-digit number come back whole, such as nist sp 800-207.
  • Job title and location (output_object=job): the job's location (both parsing models) and its name (Quicksilver) are the place and the job title written first in the posting, so a posting that mentions its head office or the manager the role reports to further down keeps its own city and title.
  • Contact details (output_object=profile): the first_name, last_name, email, phone and location of the profile's info are the ones written first in the text, usually the candidate's own, not those of a referee named further down.
  • Job sections with Atlas (output_object=job): a posting whose list bullets sit alone on their lines, or that holds long runs of empty lines, comes back with its summary, requirements, responsibilities and other sections filled, and faster.
  • Copied sections with Atlas (output_object=job): the sections copied from the posting keep every character where a line breaks mid-sentence, and requirements more often includes the years of experience and the seniority a posting gives only in a header line, such as Senior, 15+ years.
  • Driving licences: driving_license is found more often in parsed résumés: a licence written with several classes run together, such as Permis ABC, is read as its first class.
  • Document OCR sections: POST /text/ocr returns the whole page in the sections of each of its pages: joined together, a page's sections give back its full text, and on a PDF page with a text block made only of spaces, the sections before and after it come back as two entries.

Example

curl -X POST https://api.hrflow.ai/v1/text/parsing \
  -H "X-API-KEY: <your_api_key>" \
  -H "X-USER-EMAIL: <your_email>" \
  -H "Content-Type: application/json" \
  -d '{
    "texts": ["<a job posting>"],
    "parsing_model": "quicksilver",
    "output_object": "job"
  }'

The response keeps the format documented in the reference; the job's name and location are the title and the place written first in the posting.

Links