cim_summary/backend/test-financial-extraction.js

require('dotenv').config();
const { Pool } = require('pg');
const { Anthropic } = require('@anthropic-ai/sdk');

const pool = new Pool({
  connectionString: 'postgresql://postgres:password@localhost:5432/cim_processor'
});

const anthropic = new Anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY,
});

async function testFinancialExtraction() {
  try {
    // Get the extracted text from the STAX document
    const result = await pool.query(`
      SELECT extracted_text
      FROM documents 
      WHERE id = 'b467bf28-36a1-475b-9820-aee5d767d361'
    `);
    
    if (result.rows.length === 0) {
      console.log('❌ Document not found');
      return;
    }
    
    const extractedText = result.rows[0].extracted_text;
    console.log('📄 Testing Financial Data Extraction...');
    console.log('=====================================');
    
    // Create a more specific prompt for financial data extraction
    const prompt = `You are a financial analyst extracting structured financial data from a CIM document.

IMPORTANT: Look for financial tables, charts, or structured data that shows historical financial performance.

The document contains financial data. Please extract the following information and map it to the requested format:

**LOOK FOR:**
- Revenue figures (in millions or thousands)
- EBITDA figures (in millions or thousands) 
- Financial tables with years (2023, 2024, 2025, LTM, etc.)
- Pro forma adjustments
- Historical performance data

**MAP TO THIS FORMAT:**
- FY-3: Look for the oldest year (e.g., 2022, 2023, or earliest year mentioned)
- FY-2: Look for the second oldest year (e.g., 2023, 2024)
- FY-1: Look for the most recent full year (e.g., 2024, 2025)
- LTM: Look for "LTM", "TTM", "Last Twelve Months", or most recent period

**EXTRACTED TEXT:**
${extractedText.substring(extractedText.length - 5000)} // Last 5000 characters where financial data usually appears

Please return ONLY a JSON object with this structure:
{
  "financialData": {
    "fy3": {
      "revenue": "amount or 'Not found'",
      "ebitda": "amount or 'Not found'",
      "year": "actual year found"
    },
    "fy2": {
      "revenue": "amount or 'Not found'", 
      "ebitda": "amount or 'Not found'",
      "year": "actual year found"
    },
    "fy1": {
      "revenue": "amount or 'Not found'",
      "ebitda": "amount or 'Not found'", 
      "year": "actual year found"
    },
    "ltm": {
      "revenue": "amount or 'Not found'",
      "ebitda": "amount or 'Not found'",
      "period": "LTM period found"
    }
  },
  "notes": "Any observations about the financial data found"
}`;

    const message = await anthropic.messages.create({
      model: "claude-3-5-sonnet-20241022",
      max_tokens: 2000,
      temperature: 0.1,
      system: "You are a financial analyst. Extract financial data and return ONLY valid JSON. Do not include any other text.",
      messages: [
        {
          role: "user",
          content: prompt
        }
      ]
    });

    const responseText = message.content[0].text;
    console.log('🤖 LLM Response:');
    console.log(responseText);
    
    // Try to parse the JSON response
    try {
      const parsedData = JSON.parse(responseText);
      console.log('\n✅ Parsed Financial Data:');
      console.log(JSON.stringify(parsedData, null, 2));
    } catch (parseError) {
      console.log('\n❌ Failed to parse JSON response:');
      console.log(parseError.message);
    }
    
  } catch (error) {
    console.error('❌ Error:', error.message);
  } finally {
    await pool.end();
  }
}

testFinancialExtraction();