Feedback Processor AI Agent

Overview

The Feedback Processor AI Agent is an AI-powered system that processes customer feedback based on API user instructions. It consists of two Lambda-based agents that analyze and categorize feedback while ensuring compliance with predefined guardrails.

Architecture

1. Guardrail Agent

• Deployed as: AWS Lambda function
• Function:
  • Evaluates user instructions for safety.
  • Routes messages to either the Task Executor Queue or Send Response Queue based on the is_safe attribute.

2. Task Executor Agent

• Deployed as: AWS Lambda function
• Function:
  • Executes various NLP tools to process feedback as per user instructions.
  • Plans tool sequence execution and runs tools concurrently.
  • Stores results in Agent-State DynamoDB with:
    • Primary Key: request_id
    • Global Secondary Index (GSI): feedback_id

Available NLP Tools:

• Sentiment Analysis Tool: Determines sentiment (Positive, Negative, Neutral).
• Topic Categorization Tool: Categorizes feedback into predefined topics (e.g., Product Quality, Delivery, Support).
• Keyword Contextualization Tool: Extracts relevant keywords with scores.
• Summarization Tool: Generates concise summaries and actionable insights.

3. Send Response Agent

• Deployed as: AWS Lambda function
• Function:
  • Reads from the Send Response Queue.
  • Fetches the processed feedback using request_id from DynamoDB.
  • Sends results to the WebSocket connection using connection_id.

Cloud Architecture

AWS Infrastructure

• AWS Lambda Functions: Guardrail, Task Executor, Send Response
• DynamoDB: Stores processed feedback
• SQS Queues: Task Executor Queue, Send Response Queue
• API Gateway (WebSocket): Single route (request.body.action="process")
• IAM Role: Single role with permissions for Lambda execution, DynamoDB, and SQS

Repository Structure

📂 src
 ├── 📂 lambda_functions
 │   ├── 📂 guardrail
 │   │   ├── lambda_function.py
 │   │   ├── prompts.py
 │   │   ├── tools.py
 │   ├── 📂 task_executor
 │   │   ├── lambda_function.py
 │   │   ├── prompts.py
 │   │   ├── tools.py
 │   ├── 📂 send_response
 │   │   ├── lambda_function.py
 │   │   ├── prompts.py
 │   │   ├── tools.py
 ├── 📂 experiments
 ├── .env
 ├── requirements.txt

API Testing

WebSocket API Endpoint:

wss://7q3r72lbzc.execute-api.ap-south-1.amazonaws.com/production/

Sample API Request:

{
    "action":"process",
    "instructions":"Categorize the topic and analyse the sentiment",
    "feedback_id":"12345",
    "feedback_text":"Amazing product but checkout process is difficult",
    "debug": true
}

Local Setup Instructions

1. Create Virtual Environment & Install Dependencies

uv venv

uv sync

2. Set Environment Variables

• Define required variables in .env file.

Future Improvements

• Implement more robust instruction validation for Guardrail Agent.
• Introduce additional NLP tools for enhanced feedback processing.
• Optimize concurrency management for better performance.

---

Feel free to contribute and enhance this project! 🚀