src/shared/llm/prompts.ts

/**
 * Production-grade prompt library.
 * 
 * Design principles (Google/Anthropic standard):
 * 1. Chain-of-thought: Force reasoning before conclusion
 * 2. Few-shot examples: 2-3 examples for each prompt
 * 3. Structured output: Exact JSON schema specified
 * 4. Grounding instruction: "Only state what evidence supports"
 * 5. Anti-hallucination: "Write UNKNOWN if data not provided"
 * 6. Token-efficient: No verbose instructions, no repetition
 */

// ─── SYSTEM PROMPTS ──────────────────────────────────────────

export const SYSTEM_PROMPTS = {
  PROFILER: `You are a business analyst for an AI automation agency.
Your job: analyze a company and identify WHERE our AI services can help them.

CRITICAL RULES:
- Only state facts supported by the provided evidence
- Write "UNKNOWN" for anything not in the data — NEVER guess
- Your analysis determines whether a real salesperson contacts this company
- Wrong analysis = wasted human time = unacceptable
- Think step by step before concluding`,

  SCORER: `You are a lead qualification engine.
Your job: extract SIGNALS from company data. You do NOT compute the final score.
The system computes scores deterministically from your signal extraction.

CRITICAL RULES:
- Extract only what the evidence supports
- For each signal, cite which piece of evidence supports it
- If evidence is weak or missing, say so honestly
- Output ONLY the structured JSON requested`,

  EMAIL_CLASSIFIER: `You are a B2B email quality analyst.
Your job: determine if a specific email address reaches a decision-maker.
Consider company size, industry, and the email prefix meaning in context.

CRITICAL RULES:
- Small company (<20 people): admin@, operations@, office@ likely reaches owner
- Large company (200+): same prefixes likely reach departments, not individuals
- NEVER assume — reason from the evidence provided
- When uncertain, err on the side of KEEPING the email (mark confidence low)`,

  PAIN_DETECTOR: `You are an operations efficiency analyst.
Your job: identify operational pain points in a company that AI automation can solve.
You are NOT looking for companies that already use AI.
You ARE looking for companies with manual, repetitive, or inefficient processes.

CRITICAL RULES:
- A phone number on homepage = manual call handling (pain point)
- "Book by phone" = no online scheduling (pain point)  
- No chatbot visible = manual customer interaction (pain point)
- Small staff + many services = overworked team (pain point)
- These are REAL signals, not guesses`,
} as const;


// ─── PROFILING PROMPT ────────────────────────────────────────

export function buildProfilePrompt(companyData: {
  name: string;
  industry: string;
  employee_count: number | null;
  description: string;
  website_text: string;
  tech_stack: string[];
  job_postings: string[];
  ai_job_count: number;
  linkedin_description: string;
  pain_signals: string[];
  service_match: string | null;
}): string {
  return `ANALYZE THIS COMPANY:

Name: ${companyData.name}
Industry: ${companyData.industry || "UNKNOWN"}
Employees: ${companyData.employee_count ?? "UNKNOWN"}
Description: ${companyData.description || "NONE PROVIDED"}

Website excerpt (first 600 chars):
${(companyData.website_text || "").slice(0, 600)}

LinkedIn description:
${companyData.linkedin_description || "NONE"}

Tech stack detected: ${companyData.tech_stack.length ? companyData.tech_stack.join(", ") : "NONE DETECTED"}
Job postings mentioning AI/automation: ${companyData.ai_job_count}
Pain signals detected: ${companyData.pain_signals.length ? companyData.pain_signals.join(", ") : "NONE"}
Service match suggestion: ${companyData.service_match || "NONE"}

STEP-BY-STEP ANALYSIS:

Step 1: What does this company actually DO? (2 sentences, facts only)
Step 2: What are their likely daily operational challenges? (based on industry + size)
Step 3: What specific AI automation would save them time/money? (be specific)
Step 4: Who in this organization would approve buying this service?
Step 5: What outreach angle would resonate with this specific person?

After reasoning through steps 1-5, output this JSON:
{
  "profile_summary": "2-3 factual sentences about what this company does",
  "pain_points": ["specific pain 1", "specific pain 2"],
  "ai_use_case": "The single most compelling AI use case for them",
  "ai_readiness": "low|medium|high",
  "decision_maker_reasoning": "Who likely makes purchasing decisions and why",
  "outreach_angle": "One specific sentence — the hook for first contact",
  "confidence": 0.0,
  "evidence_used": ["list which data points you relied on"],
  "evidence_missing": ["list what data you wished you had"]
}

EXAMPLE 1 (dental clinic, 6 employees):
{
  "profile_summary": "ABC Dental is a 6-person dental practice in Houston offering general and cosmetic dentistry. They display their phone number prominently and use a basic contact form for appointments.",
  "pain_points": ["Manual phone-based appointment scheduling during business hours only", "No after-hours patient communication capability"],
  "ai_use_case": "AI receptionist to handle appointment booking, reminders, and after-hours calls",
  "ai_readiness": "low",
  "decision_maker_reasoning": "Practice owner (Dr. Smith, DDS) makes all purchasing decisions. Small practice = owner controls budget directly.",
  "outreach_angle": "Stop losing patients to voicemail — our AI receptionist books appointments 24/7, even when your front desk is closed",
  "confidence": 0.82,
  "evidence_used": ["phone number on homepage", "contact form only", "6 staff listed", "no chatbot detected"],
  "evidence_missing": ["annual revenue", "number of daily calls", "current scheduling software"]
}

EXAMPLE 2 (manufacturing company, 150 employees):
{
  "profile_summary": "XYZ Manufacturing is a UK-based manufacturer of industrial valves with 150 employees. They use SAP for ERP and are hiring a Data Analyst, suggesting manual reporting pain.",
  "pain_points": ["Manual data extraction from legacy SAP system", "Production reporting requires manual spreadsheet compilation"],
  "ai_use_case": "Automated reporting pipeline that extracts SAP data and generates dashboards without manual intervention",
  "ai_readiness": "medium",
  "decision_maker_reasoning": "Operations Director (found on LinkedIn) manages the data team and would champion this internally. CTO signs off on tech purchases.",
  "outreach_angle": "Your Data Analyst job posting tells us you're drowning in manual SAP reports — we automate that entirely",
  "confidence": 0.88,
  "evidence_used": ["SAP detected in tech stack", "Data Analyst job posting", "150 employees", "manufacturing industry"],
  "evidence_missing": ["specific SAP modules used", "current reporting frequency"]
}`;
}


// ─── SIGNAL EXTRACTION PROMPT (for scoring) ──────────────────

export function buildSignalExtractionPrompt(companyData: {
  name: string;
  industry: string;
  employee_count: number | null;
  tech_stack: string[];
  ai_job_count: number;
  pain_signals: string[];
  service_match: string | null;
  has_verified_email: boolean;
  has_linkedin: boolean;
  has_social: boolean;
  growth_signals_count: number;
  website_active: boolean;
}): string {
  return `EXTRACT SIGNALS for lead scoring. Do not compute a score — just identify signals.

Company: ${companyData.name}
Industry: ${companyData.industry || "UNKNOWN"}
Employees: ${companyData.employee_count ?? "UNKNOWN"}
Tech stack: ${companyData.tech_stack.join(", ") || "NONE"}
AI/automation job postings: ${companyData.ai_job_count}
Pain signals detected: ${companyData.pain_signals.join(", ") || "NONE"}
Service match: ${companyData.service_match || "NONE"}
Has verified email: ${companyData.has_verified_email}
Has personal LinkedIn: ${companyData.has_linkedin}
Has social profiles: ${companyData.has_social}
Growth signals count: ${companyData.growth_signals_count}
Website recently active: ${companyData.website_active}

Output JSON:
{
  "company_fit_signals": {
    "industry_match": true|false,
    "size_appropriate": true|false,
    "evidence": "why"
  },
  "ai_readiness_signals": {
    "level": "none|low|medium|high",
    "tech_stack_relevant": true|false,
    "ai_jobs_present": true|false,
    "evidence": "why"
  },
  "service_match_signals": {
    "matched": true|false,
    "service_name": "which service fits",
    "pain_count": 0,
    "evidence": "which pain signals"
  },
  "contact_quality_signals": {
    "email_verified": true|false,
    "linkedin_found": true|false,
    "decision_maker_identified": true|false
  },
  "timing_signals": {
    "actively_growing": true|false,
    "recently_active": true|false,
    "evidence": "what suggests good timing"
  },
  "confidence": 0.0
}`;
}


// ─── EMAIL CLASSIFICATION PROMPT ─────────────────────────────

export function buildEmailClassifyPrompt(data: {
  email: string;
  company_name: string;
  company_size: number | null;
  industry: string;
  website_snippet: string;
}): string {
  return `CLASSIFY this email address for B2B outreach viability.

Email: ${data.email}
Company: ${data.company_name}
Size: ${data.company_size ?? "UNKNOWN"} employees
Industry: ${data.industry || "UNKNOWN"}
Website excerpt: ${(data.website_snippet || "").slice(0, 300)}

Does "${data.email}" likely reach a person with purchasing authority?

Consider:
- Email prefix meaning in context of this company size
- "${data.email.split("@")[0]}@" at a ${data.company_size ?? "unknown"}-person ${data.industry} company
- Small companies: admin/operations/office = often the owner
- Large companies: admin/operations = departments, not individuals

Output JSON:
{
  "keep": true|false,
  "confidence": 0.0,
  "likely_reaches": "who this email probably reaches",
  "reason": "one line why keep or reject"
}`;
}


// ─── PAIN SIGNAL DETECTION PROMPT ────────────────────────────

export function buildPainDetectionPrompt(data: {
  company_name: string;
  industry: string;
  employee_count: number | null;
  website_text: string;
  page_elements: string[];  // ['phone_number', 'contact_form', 'no_chatbot', etc.]
}): string {
  return `DETECT operational inefficiency signals for this company.

Company: ${data.company_name}
Industry: ${data.industry || "UNKNOWN"}
Size: ${data.employee_count ?? "UNKNOWN"} employees

Website text (excerpt):
${(data.website_text || "").slice(0, 500)}

Page elements detected:
${data.page_elements.join("\n")}

IMPORTANT: You are NOT looking for AI signals. You are looking for MANUAL PROCESS signals.
A phone number on a homepage IS a signal (manual call handling).
A "Book by Phone" button IS a signal (no online scheduling).
No live chat IS a signal (no automated customer interaction).

Output JSON:
{
  "pain_signals": [
    {"signal": "what you detected", "evidence": "where on page", "severity": "low|medium|high"}
  ],
  "service_match": "which AI service best fits: AI Receptionist|AI Customer Support|AI Data Processing|AI Sales Automation|AI Workflow Automation|NONE",
  "match_confidence": 0.0,
  "reasoning": "one paragraph explaining your analysis"
}`;
}