🤖 Autonomous E-Commerce Diagnostic & Remediation Agent

An "Agentic" serverless solution powered by Amazon Bedrock that autonomously diagnoses, remediates, and verifies sellability issues in e-commerce systems.

🚀 Overview

In large-scale e-commerce, synchronization between upstream systems (Inventory Service, Pricing Engine, PIM Service) and the web database (DynamoDB) often fails due to transient race conditions or data inconsistencies. This project implements a self-healing diagnostic agent that business users can query in natural language to repair product data discrepancies automatically.

✨ Key Features

• ⛓️ Agentic Chain-of-Thought: The agent uses an autonomous reasoning loop (Diagnose → Investigate → Remediate → Verify) instead of simple scripts.
• 🛠️ Self-Healing Auto-Remediation: Automatically detects transient "Database Lock" errors (race conditions) in the Dead Letter Queue (DLQ) and triggers a manual override sync.
• 📚 Agentic RAG (Mock Knowledge Base): Features a queryTroubleshootingGuide tool that allows the agent to "research" technical error codes manually before deciding on a resolution.
• ✅ Closed-Loop Verification: After performing a fix, the agent proactively re-checks the web state to ensure the product is truly "Sellable" before notifying the user.

---

🏗️ Architecture

graph TD
    User([Business User]) -->|Natural Language| Agent[Amazon Bedrock Agent]
    Agent -->|Reasoning Loop| Executor[AWS Lambda Executor]
    
    subgraph "Diagnostic Tools"
        Executor -->|Check Web DB| Dynamo[(Web DynamoDB)]
        Executor -->|Check Upstream| INV[(Inventory Service)]
        Executor -->|Check Upstream| PRCE[(Pricing Engine)]
        Executor -->|Check Metadata| PIM[(PIM Service)]
        Executor -->|Check Queue| SQS[SQS/DLQ Health]
    end
    
    subgraph "Remediation & RAG"
        Executor -->|Trigger Sync| SyncAPI[Sync API]
        Executor -->|Consult Guide| RAG[Troubleshooting Guide]
    end

Loading

---

🛠️ Tech Stack

• AI Orchestration: Amazon Bedrock Agents (Claude 3 Sonnet)
• Compute: AWS Lambda (TypeScript / Node.js 18.x)
• Infrastructure: Serverless Framework (Infrastructure as Code)
• Security: IAM Roles & Action Group Permissions

---

📂 Project Structure

• src/handlers/executor.ts: The core "Action Group" executor where the diagnostic and remediation logic lives.
• serverless.yml: Defines the Bedrock Agent, Action Groups, and Lambda permissions.
• src/handlers/chat.ts: A helper function for programmatically interacting with the agent.

---

🚀 Getting Started

⚡ Instant Local Chat

New in Project 1: Test the agent logic directly from your terminal!

1. Copy .env.example to .env and fill in your AGENT_ID and AGENT_ALIAS_ID.
2. Install & Chat:

   npm install
   npm run chat

🧪 Automated Evaluation

Project 1 now includes a built-in evaluation suite to measure agent accuracy:

1. Ensure your agent is deployed (sls deploy) and your AGENT_ID is set in .env.
2. Run the suite:

   npm run eval

Note: This suite tests your live Bedrock Agent deployment to verify its multi-step diagnostic logic.

Prerequisites

• Active AWS Account with Amazon Bedrock Model Access (Claude 3 Sonnet).
• Serverless Framework installed (npm install -g serverless).
• AWS CLI configured with necessary permissions.

Deployment

1. Clone the repository:

   git clone [your-repo-url]
   cd bedrock-full-reconciliation

2. Install dependencies:

   npm install

3. Deploy to AWS:

   sls deploy --stage dev

---

📖 Example Scenario: "Self-Healing"

User: "Why is product prod_9982 not showing up on the site?"

1. Diagnosis: Agent calls checkWebDatabase (Result: Status is NOT_SELLABLE due to 0 inventory).
2. Comparison: Agent calls checkInventoryService (Result: Core system has 150 units in stock).
3. Root Cause Analysis: Sensing a discrepancy, the agent calls checkDeadLetterQueue and finds a ConsumerDatabaseTimeoutException (Race Condition).
4. Research: Agent calls queryTroubleshootingGuide and learns that an autonomous syncSystem override will resolve this lock contention.
5. Remediation: Agent calls syncSystem(system="inventory") to force-refresh the web database.
6. Verification: Agent re-runs checkWebDatabase and confirms the site is now SELLABLE.
7. Resolution: Agent informs the user the diagnostic and repair are complete!

---

✍️ Author

Palamkunnel Sujith - AI & Serverless Architect

• LinkedIn: [https://www.linkedin.com/in/sujithpvarghese/]

⚖️ License

Distributed under the MIT License.