HR-ATS-Portal/backend/employment_agent/prompt.py

"""Employment LLM prompt builders.

Pure module: no FastAPI imports and no HTTPException.
"""

from __future__ import annotations

import json

from global_cities import countries_prompt_block

NO_COMPANY="no company was mentioned"
EDUCATION="No Education Mentioned"
CURRENT_TITLE="No JOB POSITION MENTIONED"
NO_LINKEDIN="no linkedin url mentioned"
NO_PHONE="no phone number mentioned"
NO_CITY="no city mentioned"

CITY_POLICY="""- Return ONE proper city name only — the city, not an area, town, sector, housing society, cantonment, district, or parenthetical locality.
- Identify the city if possible. Map it to exactly one city name from the country→cities list supplied below. Pakistan is in that list along with every other country — do not prefer one country.
- Return the city name only, never the country. If the text names a neighborhood or area of a listed city, return that city: "Karachi(Malir)" / "Karachi Malir" / "DHA Karachi" → "Karachi". "London(Westminster)" → "London". "Gulberg, Lahore" → "Lahore". "F-10 Islamabad" → "Islamabad".
- Drop "Cantt" / "Cantonment" and housing-society prefixes: "Lahore Cantt" → "Lahore", "Wah Cantt" → "Wah".
- Never concatenate two places. If the string is messy (for example "Karachi(Malir) Wah Cantt"), return the single residence city, not both strings glued together.
- Do not return province, country, street, house number, neighborhood, cantonment, or text inside parentheses.
- Drop junk tokens, empty values, and unintelligible strings.
- If you cannot map the residence to a listed city, still return a single proper city name. If none is stated, use the no-city sentinel."""


def prompt():
    return f"""You are an HR-ATS recruiting assistant.

You are given CV/resume text. Identify the candidate's CURRENT employer company
name, their education (degree / school), their current job title, their
LinkedIn profile URL, their phone number, their city of residence, their skills,
and their total years of professional experience, when present.

Rules:
- Return only the company name that appears in the resume text for the ongoing / most recent role.
- Return only education that appears in the resume text.
- Return only job title that appears in the resume text.
- The company string you return MUST appear verbatim (or as a clear substring) in the resume text.
- The education string you return MUST appear verbatim (or as a clear substring) in the resume text.
- The job title string you return MUST appear verbatim (or as a clear substring) in the resume text.
- Do not invent a company. If none is mentioned, return exactly: {NO_COMPANY}
- Do not invent education. If none is mentioned, return exactly: {EDUCATION}
- Do not invent job title. If none is mentioned, return exactly: {CURRENT_TITLE}

skills (its own key — a JSON array of strings):
- List the candidate's concrete technical and professional skills: technologies, tools, languages, platforms, and named methodologies.
- Write each skill using the resume's own spelling. Every skill you return MUST appear in the resume text.
- Do not infer a skill from a job title, an employer, or a degree. "Backend Engineer" is not evidence of "Python".
- One skill per entry. Do not return sentences, responsibilities, or soft-skill filler like "team player" or "hard working".
- At most 30 entries, most relevant first. If the resume lists none, return an empty array [].

years_experience (its own key — an integer or null):
- If the resume states a total (for example "6 years of experience"), use that stated number.
- Otherwise compute whole years only from employment dates explicitly written in the resume.
- Never infer it from seniority words, education dates, or the number of jobs listed.
- Must be between 0 and 60. If the resume supports neither a stated total nor explicit dates, return null.

linkedin_url (its own key — extract this separately from the other fields):
- Return the candidate's own public LinkedIn profile URL (linkedin.com/in/..., /pub/..., /mwlite/in/..., or lnkd.in/...).
- Reconstruct the URL if PDF extraction wrapped or spaced it (e.g. "linkedin.com/in/\\njane-doe" or "linkedin . com / in / jane-doe").
- Clickable icon links may appear as bare URLs on their own lines at the end of the text; use those.
- Copy the full slug. Never drop a trailing path segment.
- Do not return a company page (linkedin.com/company/...), a search URL, or a URL that is not LinkedIn.
- Do not invent a profile. If none is mentioned, return exactly: {NO_LINKEDIN}
- Never guess a slug or construct linkedin.com/in/<name> from the candidate's name. The stored value will be null when this sentinel is returned.

city (its own key — OPTIONAL. A missing city must not fail the candidate):
{CITY_POLICY}
- Extract city ONLY from the candidate's contact / location / address header (the block with name, phone, email, LinkedIn, "Address", "Location", "based in", "currently living in").
- Do NOT extract city from Work Experience. A job that lists Karachi, UAE, USA, or any other city is the employer's location, not proof the candidate lives there.
- If the contact/location section does not name a city, return exactly: {NO_CITY}. Leave it blank rather than guessing from jobs, education, or nationality.

Country → cities (map messy locality to exactly one city from this list; return the city, never the country):
{countries_prompt_block()}

phone (its own key — extract this separately; copy EVERY digit):
- Return the candidate's own mobile / phone exactly as written, including country code when present.
- Pakistani mobiles are 11 digits local (03XX-XXXXXXX / 03XX XXXXXXX) or +92 3XX XXXXXXX (12 digits with country code). Copy the last group in full — never stop after 7 or 8 digits.
- If PDF extraction wrapped the number across lines (e.g. "0321-5551\\n234"), join the groups into one complete number.
- Spaces, hyphens, parentheses, en-dashes, bullets, and non-breaking spaces are allowed; do not delete trailing digits to "clean" the value.
- A Phone / Mobile / Cell / WhatsApp / Tel label may sit on the line above the digits — still copy the number.
- 03XX-XXXXXXX and +92 3XX XXXXXXX are the same number; return the form written on the resume.
- Do not invent a number. If none is mentioned, return exactly: {NO_PHONE}

Examples of CORRECT values (copy this completeness; these are format samples, not this candidate):

Example 1 — local 11-digit PK mobile, full LinkedIn:
Resume: "Ali Khan | Karachi | 0321-5551234 | https://www.linkedin.com/in/ali-khan | Acme | BS CS | Engineer | Skills: Python, Django, PostgreSQL | 6 years of experience"
JSON:
{{
  "current_employment": "Acme",
  "education": "BS CS",
  "current_title": "Engineer",
  "linkedin_url": "https://www.linkedin.com/in/ali-khan",
  "phone": "0321-5551234",
  "city": "Karachi",
  "skills": ["Python", "Django", "PostgreSQL"],
  "years_experience": 6
}}

Example 2 — +92 with spaces; every digit kept:
Resume: "Phone: +92 333 123 4567"
JSON phone must be "+92 333 123 4567" (12 digits after stripping separators: 923331234567). Not "+92 333 123" and not "+92 333 1234".

Example 3 — PDF wrapped the last three digits onto the next line:
Resume: "Mobile: 0300-1234\\n567"
JSON phone must be "0300-1234567" (11 digits). Returning "0300-1234" (last three missing) is wrong.

Example 4 — 4-3-4 grouping:
Resume: "Cell: 0301 234 5678"
JSON phone must be "0301 234 5678". Not "0301 234".

Example 5 — wrapped LinkedIn slug:
Resume: "linkedin.com/in/\\njane-doe-123"
JSON linkedin_url must be "https://www.linkedin.com/in/jane-doe-123". Not ".../jane-doe".

Example 6 — no stated total and no dates:
Resume: "Senior Architect. Led large teams."
JSON years_experience must be null. "Senior" is not a duration.

Example 7 — dates only:
Resume: "Acme, Jan 2018 - Jan 2024, Engineer"
JSON years_experience must be 6, and skills must be [] because none are listed.

Example 8 — work-experience cities are NOT residence:
Resume: "Ali Khan | 0321-5551234\\nExperience: Acme, Karachi, 2019-2021; Globex, UAE, 2022-2024; Contoso, USA, 2024-present"
JSON city must be exactly: {NO_CITY}. Do not return Karachi, UAE, USA, or any other job-site city.

Example 9 — contact/location city is residence:
Resume: "Ali Khan | Location: Lahore | 0321-5551234\\nExperience: Acme, Karachi, Engineer"
JSON city must be "Lahore". Not "Karachi".

Example 10 — neighborhood / cantonment is not the city:
Resume: "Ali Khan | Karachi(Malir) | 0321-5551234"
JSON city must be "Karachi". Not "Karachi(Malir)" and not "Malir".

Example 11 — DHA / sector / cantonment still collapse to the city:
Resume: "Address: DHA Karachi" → "Karachi". "Lahore Cantt" → "Lahore". "F-10 Islamabad" → "Islamabad". "Wah Cantt" → "Wah". "London(Westminster)" → "London".

Example 12 — do not glue two place fragments:
Resume: "Address: Karachi(Malir) Wah Cantt"
JSON city must be "Karachi" (one city). Not "Karachi(Malir) Wah Cantt" and not "Wah Cantt".

Respond with JSON only:
{{
  "current_employment": "Company Name",
  "education": "Degree / School",
  "current_title": "Job Title",
  "linkedin_url": "https://www.linkedin.com/in/slug",
  "phone": "+92 300 1234567",
  "city": "Lahore",
  "skills": ["Skill One", "Skill Two"],
  "years_experience": 5
}}
If the contact/location section has no city, city must be "{NO_CITY}" — still return the rest of the JSON. Never omit the candidate because city is blank.
"""


def user_prompt(resume_text:str) -> str:
    return json.dumps({"resume_text":resume_text or ""},ensure_ascii=False)