scripts/verify_weak_llm_recovery.py

"""Test ATS recovery when the LLM produces a weak/sparse v2 output (mimics real production)."""
import os, sys, io, pdfplumber
sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding='utf-8')


def read_pdf(p):
    text = ''
    with pdfplumber.open(p) as pdf:
        for page in pdf.pages:
            t = page.extract_text()
            if t:
                text += t + '\n'
    return text


orig = read_pdf(r'C:\Users\Nxtwave\Desktop\resume\Saiteja_Tirunagari_Resume A 26 - Copy.pdf')

from src.resume_customizer import ResumeCustomizer, _read_docx_text
from src.ats_scorer import score_resume
from src.resume_parser import ResumeParser

rc = ResumeCustomizer.__new__(ResumeCustomizer)
rc.resume_text = orig
rc.output_dir = 'data/output/resumes/_weak_llm_test'
os.makedirs(rc.output_dir, exist_ok=True)
rc.fast_model_cfg = None
parser = ResumeParser.__new__(ResumeParser)
parser.pdf_path = ''
contact = parser.get_contact_info(orig)


# Simulate a WEAK LLM that follows v1 schema (older model) — most production runs
# look like this: just summary + a flat experience_bullets dict, no rewritten_bullets
def weak_llm_v1(jd_label, summary_text):
    return {
        "professional_summary": summary_text,
        # v1 schema — backwards compat path
        "core_competencies": ["Product Strategy", "Roadmap Planning", "A/B Testing"],
        "experience_bullets": {
            "internal_product_manager": [
                "Led NIAT Application Portal revamp",
                "Built AI chatbot funnel",
                "Drove payment conversion optimization",
            ],
        },
        "key_achievements": ["141,269 verified leads", "+35.87 pp payment lift"],
    }


jds = [
    ("Airtel PM", "tests/fixtures/jds/airtel_pm.txt"),
    ("Sumo Logic PM", "tests/fixtures/jds/sumo_logic_pm.txt"),
    ("EdgeVerve PM", "tests/fixtures/jds/edgeverve_pm.txt"),
    ("Aditya Birla APM", "tests/fixtures/jds/aditya_birla_apm.txt"),
]

print('━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━')
print(f'{"JD":<22} {"Before inj":<14} {"After inj":<14} {"Missing":<6}')
print('━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━')

for label, jd_path in jds:
    with open(jd_path, encoding='utf-8') as f:
        jd = f.read()

    # Weak LLM with a minimal summary
    weak_summary = (
        f"Product Manager with 5+ years of experience driving 0-1 product "
        f"development in fast-paced startups."
    )
    cust = weak_llm_v1(label, weak_summary)

    filepath = os.path.join(rc.output_dir, f'{label.replace(" ", "_")}.docx')
    job = {'title': 'PM', 'company': label, 'relevance_score': 7}
    rc._write_docx(filepath, job, cust, contact)

    # Score before injection
    r_before = score_resume(_read_docx_text(filepath), jd)

    # Apply injection
    rc._inject_missing_keywords(filepath, jd)

    # Score after injection
    r_after = score_resume(_read_docx_text(filepath), jd)

    print(f'{label:<22} {r_before["ats_score"]:>3}/100 ({r_before["jd_match_score"]}%kw)  '
          f'{r_after["ats_score"]:>3}/100 ({r_after["jd_match_score"]}%kw)  '
          f'{len(r_after["missing_kw"]):<6}')

print('━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━')
print('\nNote: "Before inj" = weak v1-schema LLM result (typical production case)')
print('      "After inj"  = after the new aggressive keyword injection into summary')