lib/company_tier.py

"""
lib/company_tier.py  — PRINCIPAL-LEVEL UPGRADE

Company quality is the SINGLE BIGGEST missing signal in the original system.
The JD explicitly and repeatedly emphasizes "product companies (not pure services)"
and this signal was completely absent from scoring.

This module provides a tiered company quality score based on:
  1. Known global tech titans (FAANG+)
  2. Major product companies (Indian & global)
  3. AI/ML-native product companies
  4. Known services/consulting firms (negative signal)
  5. Unknown companies (neutral)

Design rationale:
- The JD says "If your career has been entirely at consulting firms... we won't move forward"
- But the disqualifier_penalty only fires when ALL roles are at consulting firms
- A candidate currently at Genpact AI with one prior startup role escapes the penalty
- Company quality as a continuous signal captures this gradient
- Also differentiates Apple/Meta (Tier 1) from a random startup (Tier 3)
"""
from __future__ import annotations

# Tier 1: Global tech titans — the JD's "Google or Meta" benchmark
GLOBAL_TECH_TIER1 = [
    "google", "alphabet", "deepmind",
    "meta", "facebook",
    "microsoft",
    "apple",
    "amazon",
    "netflix",
    "nvidia",
    "openai", "anthropic",
    "uber", "airbnb",
    "spotify", "linkedin",
    "bloomberg",
    "palantir", "databricks", "snowflake", "confluent",
]

# Tier 2: Major Indian & global product companies
# These are companies the JD would recognize as "product companies"
PRODUCT_TIER2 = [
    "flipkart", "razorpay", "zomato", "swiggy", "ola",
    "phonepe", "cred", "paytm", "nykaa", "dream11",
    "meesho", "cult.fit", "cure.fit",
    "inmobi", "myntra", "bigbasket",
    "twitter", "stripe", "grab", "foodpanda",
    "booking.com", "adobe", "salesforce", "service now", "servicenow",
    "atlassian", "shopify", "dropbox", "slack",
    "samsung", "intel", "ibm research",
    "baidu", "alibaba", "tencent", "bytedance",
    "stripe",
]

# Tier 3: AI/ML-native product companies (smaller but genuine product)
AI_ML_PRODUCT_TIER3 = [
    "sarvam ai", "sarvam",
    "niramai", "niramai health",
    "rephrase.ai", "rephrase",
    "haptik", "haptik ai",
    "krutrim", "krutrim ai",
    "mad street den",
    "locobuzz",
    "verloop", "verloop.io",
    "obvious.ai", "obvious",
    "fluid ai",
    "basis.ai",
    "threado",
    "electric ai", "electric",
    "kore.ai", "kore",
    "aible", "h2o.ai", "datarobot",
    "weights & biases", "wandb",
    "weaviate", "pinecone", "qdrant",
    "anyscale", "modal",
    "langchain", "llamaindex",
    "cohere", "ai21 labs",
    "stability ai", "midjourney",
    "scale ai", "labelbox",
    "snorkel", "cleanlab",
    "yellow.ai", "yellow messenger",
    "wysa", "wysa ai",
    "saarthi.ai", "saarthi",
    "agilet", "agilyx",
    "nios", "nios lab",
]

# Tier 4: Services companies disguised as AI (lower than neutral)
SERVICES_DISGUISED = [
    "genpact", "genpact ai", "genpact digital",
    "tcs", "tata consultancy", "infosys", "wipro",
    "accenture", "cognizant", "capgemini",
    "hcl", "tech mahindra", "mindtree", "lti",
    "mphasis", "hexaware", "zensar", "birlasoft",
    "niit", "cyient", "mastek", "sonata software",
    "persistent systems", "persistent",
    "wns", "firstsource", "genpact",
]

# Edtech / non-AI product (neutral to slightly positive)
EDTECH_COMPANIES = [
    "vedantu", "byju", "unacademy", "upgrad",
    "physics wallah", "pw", "eros now",
]

# HR-tech / marketplace (JD nice-to-have)
HR_MARKETPLACE = [
    "redrob", "naukri", "linkedin", "indeed",
    "glassdoor", "hire" "zee", "foundit",
]


def company_quality_score(company: str) -> float:
    """
    Returns a company quality score in [0.20, 1.00].

    Scoring:
      1.00  -- Global tech titans (FAANG+)
      0.90  -- Major product companies (Flipkart, Razorpay, etc.)
      0.80  -- AI/ML-native product companies
      0.65  -- Edtech / marketplace companies (JD nice-to-have domain)
      0.50  -- Unknown company (neutral default)
      0.30  -- Known services/consulting (even if branded as "AI")
    """
    if not company:
        return 0.50
    c = company.lower().strip()

    # Check in reverse tier order (most specific first)
    for name in SERVICES_DISGUISED:
        if name in c:
            return 0.30

    for name in GLOBAL_TECH_TIER1:
        if name in c:
            return 1.00

    for name in PRODUCT_TIER2:
        if name in c:
            return 0.90

    for name in HR_MARKETPLACE:
        if name in c:
            return 0.70

    for name in AI_ML_PRODUCT_TIER3:
        if name in c:
            return 0.80

    for name in EDTECH_COMPANIES:
        if name in c:
            return 0.55

    return 0.50  # Unknown — neutral


def company_tier_label(company: str) -> str:
    """Return human-readable tier label for reasoning."""
    score = company_quality_score(company)
    if score >= 0.95:
        return "global tech titan"
    if score >= 0.85:
        return "major product company"
    if score >= 0.75:
        return "AI/ML product company"
    if score >= 0.60:
        return "domain-relevant product company"
    if score <= 0.35:
        return "services/consulting firm"
    return "company"