templates.py

"""LLM prompt templates for resume processing."""

# Language code to full name mapping
LANGUAGE_NAMES = {
    "en": "English",
    "es": "Spanish",
    "zh": "Chinese (Simplified)",
    "ja": "Japanese",
    "pt": "Brazilian Portuguese",
}


def get_language_name(code: str) -> str:
    """Get full language name from code."""
    return LANGUAGE_NAMES.get(code, "English")


# Schema with example values - used for prompts to show LLM expected format
RESUME_SCHEMA_EXAMPLE = """{
  "personalInfo": {
    "name": "John Doe",
    "title": "Software Engineer",
    "email": "john@example.com",
    "phone": "+1-555-0100",
    "location": "San Francisco, CA",
    "website": "https://johndoe.dev",
    "linkedin": "linkedin.com/in/johndoe",
    "github": "github.com/johndoe"
  },
  "summary": "Experienced software engineer with 5+ years...",
  "workExperience": [
    {
      "id": 1,
      "title": "Senior Software Engineer",
      "company": "Tech Corp",
      "location": "San Francisco, CA",
      "years": "2020 - Present",
      "jobDescription": "Led platform modernization initiatives across customer-facing systems.",
      "description": [
        "Led development of microservices architecture",
        "Improved system performance by 40%"
      ]
    }
  ],
  "education": [
    {
      "id": 1,
      "institution": "University of California",
      "degree": "B.S. Computer Science",
      "years": "2014 - 2018",
      "description": "Graduated with honors"
    }
  ],
  "personalProjects": [
    {
      "id": 1,
      "name": "Open Source Tool",
      "role": "Creator & Maintainer",
      "years": "2021 - Present",
      "description": [
        "Built CLI tool with 1000+ GitHub stars",
        "Used by 50+ companies worldwide"
      ]
    }
  ],
  "additional": {
    "technicalSkills": ["Python", "JavaScript", "AWS", "Docker"],
    "languages": ["English (Native)", "Spanish (Conversational)"],
    "certificationsTraining": ["AWS Solutions Architect"],
    "awards": ["Employee of the Year 2022"]
  },
  "customSections": {
    "publications": {
      "sectionType": "itemList",
      "items": [
        {
          "id": 1,
          "title": "Paper Title",
          "subtitle": "Journal Name",
          "years": "2023",
          "description": ["Brief description of the publication"]
        }
      ]
    },
    "volunteer_work": {
      "sectionType": "text",
      "text": "Description of volunteer activities..."
    }
  }
}"""

PARSE_RESUME_PROMPT = """Parse this resume into JSON. Output ONLY the JSON object, no other text.

Map content to standard sections when possible. For non-standard sections (like Publications, Volunteer Work, Research, Hobbies), add them to customSections with an appropriate type.

Example output format:
{schema}

Custom section types:
- "text": Single text block (e.g., objective, statement)
- "itemList": List of items with title, subtitle, years, description (e.g., publications, research)
- "stringList": Simple list of strings (e.g., hobbies, interests)

Rules:
- Use "" for missing text fields, [] for missing arrays, null for optional fields
- Number IDs starting from 1
- Format years as "YYYY - YYYY" or "YYYY - Present"
- Use snake_case for custom section keys (e.g., "volunteer_work", "publications")
- Preserve the original section name as a descriptive key
- Normalize dates: "Jan 2020" → "2020", "2020-2021" → "2020 - 2021", "Current"/"Ongoing" → "Present"
- For ambiguous dates like "3 years experience", infer approximate years from context or use "~YYYY"
- Flag overlapping dates (concurrent roles) by preserving both, don't merge

Resume to parse:
{resume_text}"""

EXTRACT_KEYWORDS_PROMPT = """Extract job requirements as JSON. Output ONLY the JSON object, no other text.

Example format:
{{
  "required_skills": ["Python", "AWS"],
  "preferred_skills": ["Kubernetes"],
  "experience_requirements": ["5+ years"],
  "education_requirements": ["Bachelor's in CS"],
  "key_responsibilities": ["Lead team"],
  "keywords": ["microservices", "agile"],
  "experience_years": 5,
  "seniority_level": "senior"
}}

Extract numeric years (e.g., "5+ years" → 5) and infer seniority level.

Job description:
{job_description}"""

CRITICAL_TRUTHFULNESS_RULES_TEMPLATE = """CRITICAL TRUTHFULNESS RULES - NEVER VIOLATE:
1. DO NOT add any skill, tool, technology, or certification that is not explicitly mentioned in the original resume
2. DO NOT invent numeric achievements (e.g., "increased by 30%") unless they exist in original
3. DO NOT add company names, product names, or technical terms not in the original
4. DO NOT upgrade experience level (e.g., "Junior" -> "Senior")
5. DO NOT add languages, frameworks, or platforms the candidate hasn't used
6. DO NOT extend employment dates or change timelines (start/end years)
7. {rule_7}
8. Preserve factual accuracy - only use information provided by the candidate

Violation of these rules could cause serious problems for the candidate in job interviews.
"""


def _build_truthfulness_rules(rule_7: str) -> str:
    return CRITICAL_TRUTHFULNESS_RULES_TEMPLATE.format(rule_7=rule_7)


CRITICAL_TRUTHFULNESS_RULES = {
    "nudge": _build_truthfulness_rules(
        "DO NOT add new bullet points or content - only rephrase existing content"
    ),
    "keywords": _build_truthfulness_rules(
        "You may rephrase existing bullet points to include keywords, but do NOT add new bullet points"
    ),
    "full": _build_truthfulness_rules(
        "You may expand existing bullet points or add new ones that elaborate on existing work, but DO NOT invent entirely new responsibilities"
    ),
}

IMPROVE_RESUME_PROMPT_NUDGE = """Lightly nudge this resume toward the job description. Output ONLY the JSON object, no other text.

{critical_truthfulness_rules}

IMPORTANT: Generate ALL text content (summary, descriptions, skills) in {output_language}.

Rules:
- Make minimal, conservative edits only where there is a clear existing match
- Do NOT change the candidate's role, industry, or seniority level
- Do NOT introduce new tools, technologies, or certifications not already present
- Do NOT add new bullet points or sections
- Preserve original bullet count and ordering within each section
- Keep proper nouns (names, company names, locations) unchanged
- Preserve the structure of any customSections from the original resume
- Preserve original date ranges exactly - do not modify years
- If the resume is non-technical, do NOT add technical jargon
- Do NOT use em dash ("—") anywhere in the writing/output, even if it exists, remove it

Job Description:
{job_description}

Keywords to emphasize (only if already supported by resume content):
{job_keywords}

Original Resume:
{original_resume}

Output in this JSON format:
{schema}"""

IMPROVE_RESUME_PROMPT_KEYWORDS = """Enhance this resume with relevant keywords from the job description. Output ONLY the JSON object, no other text.

{critical_truthfulness_rules}

IMPORTANT: Generate ALL text content (summary, descriptions, skills) in {output_language}.

Rules:
- Strengthen alignment by weaving in relevant keywords where evidence already exists
- You may rephrase bullet points to include keyword phrasing
- Do NOT introduce new skills, tools, or certifications not in the resume
- Do NOT change role, industry, or seniority level
- Preserve the structure of any customSections from the original resume
- Preserve original date ranges exactly - do not modify years
- If resume is non-technical, keep language non-technical while still aligning keywords
- Do NOT use em dash ("—") anywhere in the writing/output, even if it exists, remove it

Job Description:
{job_description}

Keywords to emphasize:
{job_keywords}

Original Resume:
{original_resume}

Output in this JSON format:
{schema}"""

IMPROVE_RESUME_PROMPT_FULL = """Tailor this resume for the job. Output ONLY the JSON object, no other text.

{critical_truthfulness_rules}

IMPORTANT: Generate ALL text content (summary, descriptions, skills) in {output_language}.

Rules:
- Rephrase content to highlight relevant experience
- DO NOT invent new information
- Use action verbs and quantifiable achievements
- Keep proper nouns (names, company names, locations) unchanged
- Translate job titles, descriptions, and skills to {output_language}
- Preserve the structure of any customSections from the original resume
- Improve custom section content the same way as standard sections
- Preserve original date ranges exactly - do not modify years
- Calculate and emphasize total relevant experience duration when it matches requirements
- Do NOT use em dash ("—") anywhere in the writing/output, even if it exists, remove it

Job Description:
{job_description}

Keywords to emphasize:
{job_keywords}

Original Resume:
{original_resume}

Output in this JSON format:
{schema}"""

IMPROVE_PROMPT_OPTIONS = [
    {
        "id": "nudge",
        "label": "Light nudge",
        "description": "Minimal edits to better align existing experience.",
    },
    {
        "id": "keywords",
        "label": "Keyword enhance",
        "description": "Blend in relevant keywords without changing role or scope.",
    },
    {
        "id": "full",
        "label": "Full tailor",
        "description": "Comprehensive tailoring using the job description.",
    },
]

IMPROVE_RESUME_PROMPTS = {
    "nudge": IMPROVE_RESUME_PROMPT_NUDGE,
    "keywords": IMPROVE_RESUME_PROMPT_KEYWORDS,
    "full": IMPROVE_RESUME_PROMPT_FULL,
}

DEFAULT_IMPROVE_PROMPT_ID = "keywords"

# Backward-compatible alias
IMPROVE_RESUME_PROMPT = IMPROVE_RESUME_PROMPT_FULL

COVER_LETTER_PROMPT = """Write a brief cover letter for this job application.

IMPORTANT: Write in {output_language}.

Job Description:
{job_description}

Candidate Resume (JSON):
{resume_data}

Requirements:
- 100-150 words maximum
- 3-4 short paragraphs
- Opening: Reference ONE specific thing from the job description (product, tech stack, or problem they're solving) - not generic excitement about "the role"
- Middle: Pick 1-2 qualifications from resume that DIRECTLY match stated requirements - prioritize relevance over impressiveness
- Closing: Simple availability to discuss, no desperate enthusiasm
- If resume shows career transition, frame the pivot as intentional and relevant
- Extract company name from job description - do not use placeholders
- Do NOT invent information not in the resume
- Tone: Confident peer, not eager applicant
- Do NOT use em dash ("—") anywhere in the writing/output, even if it exists, remove it

Output plain text only. No JSON, no markdown formatting."""

OUTREACH_MESSAGE_PROMPT = """Generate a cold outreach message for LinkedIn or email about this job opportunity.

IMPORTANT: Write in {output_language}.

Job Description:
{job_description}

Candidate Resume (JSON):
{resume_data}

Guidelines:
- 70-100 words maximum (shorter than a cover letter)
- First sentence: Reference specific detail from job description (team, product, technical challenge) - never open with "I'm reaching out" or "I saw your posting"
- One sentence on strongest matching qualification with a concrete metric if available
- End with low-friction ask: "Worth a quick chat?" not "I'd love the opportunity to discuss"
- Tone: How you'd message a former colleague, not a stranger
- Do NOT include placeholder brackets
- Do NOT use phrases like "excited about" or "passionate about"
- Do NOT use em dash ("—") anywhere in the writing/output, even if it exists, remove it

Output plain text only. No JSON, no markdown formatting."""

# Alias for backward compatibility
RESUME_SCHEMA = RESUME_SCHEMA_EXAMPLE