Banking Services AI Agent

Agent Description:

The Banking Services AI Agent is a secure, automated system designed to provide comprehensive 24/7 banking support. It intelligently handles customer inquiries regarding account data, transaction history, and fund transfers. By connecting directly to secure banking databases via SQL, the agent provides real-time information and identifies when a scenario (such as repeated transaction failures) requires a formal support ticket.

Purpose and Components
  • Purpose: Traditional banking support is limited by long wait times, restricted service hours, and manual handling of routine inquiries, which often leads to customer frustration, errors, and potential security risks. Financial institutions require a secure, scalable solution that can interpret complex financial requests, automate routine tasks, and immediately flag high‑risk issues for escalation.

    The purpose of this agent is to provide customers with instant, secure access to financial information while automating routine banking operations. It ensures accurate data handling, delivers 24/7 support, and enables efficient escalation of critical issues—all contributing to a faster, more reliable banking experience.

  • Components:

    • Query Interpreter: Analyzes natural language messages to identify intent and extract account entities.

    • Data Retriever: The authoritative source for fetching real-time data from secure account and history tables.

    • Transaction Processor: Formats and presents financial records in a professional, customer-facing manner.

    • Feedback & Compliance Checker: Monitors for complaint intent or system anomalies to trigger formal escalations.

Supported Capabilities

  • Intent Identification: Automatically distinguishing between balance checks, transaction history requests, and escalations.

  • Structured Data Extraction: Pulling account numbers, customer IDs, and date ranges from conversational text.

  • Real-time SQL Integration: Executing precise queries against CustomerAccounts and AccountHistory tables.

  • Automated Presentation: Organizing financial data into clear, easy-to-read lists showing date, type, amount, and status.

  • Proactive Issue Handling: Detecting high-risk scenarios (for example, >3 failed transactions) and generating unique ticket IDs with resolution timelines.

LLM Used

  • OpenAI GPT_40_MINI for both the parent agent and its sub-agents.

Sub-Agents

1. Query Interpreter

  • Role:Understand User Requests.

  • Scope:Reads user messages and validates required information for downstream processing.

  • Description:Carefully interprets user intent and extracts necessary entities like account numbers. It converts this into structured JSON format without answering the user directly or accessing the database.

  • LLM Used: Default (Inherits from parent agent).

2. Data Retriever

  • Role:Data Fetcher

  • Scope:Accesses CustomerAccounts, AccountHistory, and FeedbackEscalation tables.

  • Description: A REACT agent that converts structured requests into accurate SQL queries. It returns raw results in JSON and handles error objects if accounts do not exist.

  • LLM Used: Default (Inherits from parent agent).

3. Transaction Processor

  • Role:Transaction Processor

  • Scope:Formats fund transfers, bill payments, and account updates for the user.

  • Description:Receives database outputs and formats them into organized lists. It manages user communication regarding empty results or account errors without mentioning internal technical details like SQL.

  • LLM Used: Default (Inherits from parent agent).

4. Feedback and Compliance Checker

  • Role:Issue Handler

  • Scope:Ensures transactions follow internal rules and manages the escalation workflow.

  • Description:Checks for "ESCALATION" intent (for example, "file a complaint") or system-detected triggers like excessive failed transactions. It provides a ticket confirmation or passes the previous sub-agent's response through unchanged.

  • LLM Used: Default (Inherits from parent agent).

Tools Used:
  • SQL Database Toolkit: Provides secure access to the banking_service.sql database containing account and transaction master data.
Note: For details on modifying the Tools, refer Tools Library section.
Agent Workflow Behavior Summary

  1. Input: A user sends a message: "Show my transaction history for account 98765 from last month."

  2. Intent Parsing: The Query Interpreter extracts: Account: 98765, DateRange: last month, Intent: transaction_history.

  3. Database Retrieval: The Data Retriever executes a SQL query against the AccountHistory table for the specific account and timeframe.

  4. Formatting: The Transaction Processor organizes the data into a clean, professional list for the customer.

  5. Compliance Audit: The Compliance Checker verifies if an escalation is needed. If no complaints are made and failures are low, it presents the transaction list as the final output.

Final Output: A professional summary of account data or a confirmed support ticket with a resolution timeline.

Sample Questions:
  • What is my current balance for account [Number]? Also, please escalate an issue regarding a duplicate charge I see.