💸 Financial Data Analyzer with NeuroLink (LLM + Fallback)

This project uses Juspay’s @juspay/neurolink to analyze financial transaction data using a Large Language Model (LLM). It handles LLM outputs intelligently, supports fallback logic, and ensures clean, consistent output for graphing.

---

📁 Project Structure

index.ts

🎯 Main entry point — calls the LLM to analyze financial data and produces a cleaned JSON output.

• Sends financial data to Gemini 2.5 Flash model via @juspay/neurolink.
• Expects a structured JSON response (graphType, metrics, totalAmount).
• If LLM fails or outputs invalid/messy content:
  • A robust fallback mechanism parses or reconstructs data based on original input.
• Corrects:
  • Misleading metrics (like "Total Revenue")
  • Rounding errors in percentages
  • Inconsistent totals

---

index2.ts

🧠 Enhanced version — focuses on extracting and parsing extra LLM-generated insights, along with metrics.

• Parses rich LLM explanations (observations, breakdowns, etc.)
• Extracts and displays:
  • Clean metrics
  • Graph type
  • Total amount
  • LLM commentary (e.g., "UPI is the second most used method.")
• Ideal for deeper analysis and debugging LLM behavior

---

🚀 Scripts

Command	Description
npm start	Runs the main analyzer (index.ts)
npm run start:alt	Runs the enhanced analyzer (index2.ts)

---

🧰 Requirements

• Node.js >= 18.x
• Juspay’s @juspay/neurolink SDK installed and configured
• Access to Google Vertex AI with the Gemini model enabled

---

🔒 Output Format

Both versions aim to return a valid JSON object:

{
  "graphType": "pie chart",
  "totalAmount": 300000,
  "metrics": [
    { "method": "UPI", "percentage": 23 },
    { "method": "Net banking", "percentage": 33 },
    ...
  ]
}