Autonomous Procurement System

An AI-powered procurement automation system that monitors inventory levels, identifies substitute materials, and generates professional justification emails using AWS Bedrock and Claude AI.

Architecture

The system consists of:

• EventBridge Scheduler: Triggers daily inventory checks
• Lambda Function: Core processing logic for inventory analysis and supplier evaluation
• DynamoDB: Stores raw materials inventory with technical specifications
• Bedrock (Claude AI): Generates professional procurement emails and recommendations

Features

• Automated Inventory Monitoring: Identifies items with low stock levels (< 5 days)
• Intelligent Substitute Detection: Finds alternative suppliers with identical technical specifications
• AI-Powered Communication: Generates professional emails highlighting cost savings and supply continuity
• Technical Specification Matching: Ensures substitute materials meet exact requirements
• Cost Optimization: Identifies significant price differences between suppliers

Core Components

1. Raw Materials Data Generator (generate_raw_materials.py)

Creates realistic inventory data with:

• 50 materials with SKU, supplier, price, lead time, and days on hand
• Technical specifications (density, tensile strength, grade)
• 3 pairs of interchangeable parts from different suppliers

Sample Raw Materials Data:

[
  {
    "SKU": "SKU-5895-agS",
    "SupplierName": "Baird Ltd",
    "Price": 422.22,
    "LeadTime": 6,
    "DaysOnHand": 12,
    "TechnicalSpecs": {
      "density": 2.7,
      "tensile_strength": 310,
      "grade": "6061-T6"
    }
  },
  {
    "SKU": "SKU-1066-ZeP",
    "SupplierName": "Rangel-Rodriguez",
    "Price": 116.85,
    "LeadTime": 20,
    "DaysOnHand": 16,
    "TechnicalSpecs": {
      "density": 2.7,
      "tensile_strength": 310,
      "grade": "6061-T6"
    }
  },
  {
    "SKU": "SKU-6898-DmY",
    "SupplierName": "Gibson-Middleton",
    "Price": 224.24,
    "LeadTime": 32,
    "DaysOnHand": 1,
    "TechnicalSpecs": {
      "density": 7.02,
      "tensile_strength": 476,
      "grade": "Ps58"
    }
  }
]

2. Purchase Evaluation Logic (evaluate_purchase.py)

Implements approval workflow:

• REJECTED: Technical specs don't match
• APPROVED: Specs match and price < $1000
• PENDING MANAGER: Specs match but price ≥ $1000

3. Claude AI Integration (ask_claude.py)

Connects to AWS Bedrock for:

• Natural language email generation
• Professional procurement recommendations
• Cost-benefit analysis

4. Simulation Engine (simulation.py)

Orchestrates the complete workflow:

• Identifies low-stock items
• Searches for suitable substitutes
• Generates and stores justification emails

5. Resilience Testing Suite

Comprehensive audit and testing framework:

• Resilience Auditor (resilience_auditor.py): Tests decision accuracy and substitute availability
• Compliance Validator (compliance_validator.py): Validates regulatory and business rule adherence
• Stress Tester (stress_tester.py): Simulates failure scenarios and edge cases
• Master Auditor (master_auditor.py): Orchestrates all audits and generates comprehensive reports
• Visual Report Generator (visual_report_generator.py): Creates interactive dashboards and charts
• HTML Report Generator (html_report_generator.py): Generates web-friendly audit reports

Example Output

The system generated this procurement recommendation:

Subject: Supplier Switch Proposal - SKU-5895-agS

Dear R&D Department,

I would like to propose a supplier switch for the item SKU-5895-agS, currently sourced from Baird Ltd at $422.22 per unit. With only 3 days remaining in our inventory, it is crucial to ensure a continuous supply of this item.

I have identified a suitable substitute, SKU-1066-ZeP from Rangel-Rodriguez, which meets the identical technical specifications of {'density': 2.7, 'tensile_strength': 310, 'grade': '6061-T6'}. Remarkably, the price offered by Rangel-Rodriguez is $116.85 per unit, representing a substantial cost savings of 72.3%.

By switching to Rangel-Rodriguez as our supplier for this item, we can maintain an uninterrupted supply chain while achieving significant cost reductions. I kindly request your prompt consideration of this proposal to ensure a seamless transition and capitalize on the considerable cost savings.

Thank you for your attention to this matter.

Best regards,
Procurement Team

Local Development Setup

Prerequisites

pip install faker boto3 diagrams matplotlib seaborn
brew install graphviz  # macOS only

Running the System

# Generate sample data
python3 generate_raw_materials.py

# Test purchase evaluation
python3 test_evaluate_purchase.py

# Run full simulation
python3 simulation.py

# Generate architecture diagram
python3 arch_diagram.py

# Run comprehensive resilience audit
python3 master_auditor.py

# Generate visual reports
python3 visual_report_generator.py
python3 html_report_generator.py

Resilience Testing & Compliance

The system includes comprehensive audit capabilities to ensure production readiness:

Audit Components

• Resilience Testing: Validates decision accuracy, substitute availability, and system reliability
• Compliance Validation: Ensures regulatory adherence and business rule compliance
• Stress Testing: Simulates failure scenarios, data corruption, and extreme conditions
• Visual Reporting: Generates executive dashboards and interactive reports

Sample Audit Results

Overall Health Score: 69/100
Risk Level: HIGH
Recommendation: Immediate attention required

Component Scores:
  Resilience: 50/100
  Compliance: 100/100
  Stress Resistance: 57/100

Key Findings:
  - 0% substitute coverage for low-stock items
  - 100% specification matching accuracy
  - 2 stress test failures requiring attention

Generated Reports

• master_audit_report.json: Comprehensive audit data
• audit_dashboard.png: Visual dashboard with charts and gauges
• audit_report.html: Interactive web-based report

AWS Deployment

Infrastructure Requirements

1. DynamoDB Table: raw-materials-inventory

   • Partition Key: SKU (String)
   • Attributes: SupplierName, Price, LeadTime, DaysOnHand, TechnicalSpecs

2. Lambda Function: procurement-agent

   • Runtime: Python 3.9+
   • Memory: 512 MB
   • Timeout: 5 minutes
   • Environment Variables:
     • DYNAMODB_TABLE: raw-materials-inventory
     • BEDROCK_MODEL_ID: anthropic.claude-3-sonnet-20240229-v1:0

3. EventBridge Rule: daily-procurement-check

   • Schedule: rate(1 day) or cron(0 9 * * ? *) (9 AM daily)
   • Target: Lambda function

4. IAM Role: procurement-lambda-role

   {
     "Version": "2012-10-17",
     "Statement": [
       {
         "Effect": "Allow",
         "Action": [
           "dynamodb:Scan",
           "dynamodb:GetItem",
           "dynamodb:Query"
         ],
         "Resource": "arn:aws:dynamodb:*:*:table/raw-materials-inventory"
       },
       {
         "Effect": "Allow",
         "Action": [
           "bedrock:InvokeModel"
         ],
         "Resource": "arn:aws:bedrock:*::foundation-model/anthropic.claude-3-sonnet-20240229-v1:0"
       },
       {
         "Effect": "Allow",
         "Action": [
           "ses:SendEmail",
           "sns:Publish"
         ],
         "Resource": "*"
       }
     ]
   }

Deployment Steps

1. Create DynamoDB Table:

   aws dynamodb create-table \
     --table-name raw-materials-inventory \
     --attribute-definitions AttributeName=SKU,AttributeType=S \
     --key-schema AttributeName=SKU,KeyType=HASH \
     --billing-mode PAY_PER_REQUEST

2. Load Sample Data:

   import boto3
   import json
   
   dynamodb = boto3.resource('dynamodb')
   table = dynamodb.Table('raw-materials-inventory')
   
   with open('raw_materials.json', 'r') as f:
       materials = json.load(f)
   
   for item in materials:
       table.put_item(Item=item)

3. Deploy Lambda Function:

   • Package the code with dependencies
   • Upload to Lambda with the IAM role
   • Configure environment variables

4. Create EventBridge Rule:

   aws events put-rule \
     --name daily-procurement-check \
     --schedule-expression "rate(1 day)"
   
   aws events put-targets \
     --rule daily-procurement-check \
     --targets "Id"="1","Arn"="arn:aws:lambda:region:account:function:procurement-agent"

Lambda Function Code Structure

import boto3
import json
from datetime import datetime

def lambda_handler(event, context):
    # Initialize AWS clients
    dynamodb = boto3.resource('dynamodb')
    bedrock = boto3.client('bedrock-runtime')
    
    # Scan for low-stock items
    low_stock_items = scan_inventory_for_low_stock()
    
    # Process each low-stock item
    for item in low_stock_items:
        substitute = find_substitute(item)
        if substitute:
            email = generate_email_with_bedrock(item, substitute)
            send_notification(email)
    
    return {
        'statusCode': 200,
        'body': json.dumps(f'Processed {len(low_stock_items)} items')
    }

Benefits

• Cost Savings: Automatically identifies cheaper alternatives (up to 72% savings in examples)
• Supply Continuity: Prevents stockouts through proactive monitoring
• Compliance: Ensures technical specifications are maintained
• Efficiency: Reduces manual procurement research time
• Scalability: Handles large inventories with minimal overhead
• Audit Trail: Maintains records of all recommendations and decisions
• Resilience Testing: Comprehensive audit framework ensures production readiness
• Visual Reporting: Executive dashboards provide actionable insights
• Predictive Compliance: Identifies issues before they become violations

Security Considerations

• All AWS resources use least-privilege IAM policies
• Sensitive data encrypted at rest in DynamoDB
• API calls to Bedrock are logged via CloudTrail
• Email notifications can be configured with encryption

Monitoring & Alerting

• CloudWatch metrics for Lambda execution
• DynamoDB performance monitoring
• Bedrock usage and cost tracking
• SNS notifications for critical inventory levels

This autonomous system transforms procurement from reactive to proactive, ensuring optimal supplier relationships while maintaining technical compliance and cost efficiency.