Autonomous AI Manufacturing Agent

Inspiration

The inspiration came from observing how traditional monitoring systems rely heavily on human supervision and reactive maintenance. I wanted to demonstrate how combining AWS AI/ML tools with Salesforce automation could create a closed-loop, autonomous ecosystem — one that not only detects but also acts on early warning signs.

What it does

In the manufacturing industry, even a few minutes of machine downtime can cause major production losses. This project — Autonomous Manufacturing AI Agent — was inspired by the need to proactively detect anomalies in factory equipment and automatically trigger maintenance workflows to minimize disruptions. The goal was to create an intelligent, self-operating system that detects anomalies, makes decisions using AI, and integrates with Salesforce Service Cloud to ensure rapid response and resolution.

How we built it

How I Built It The solution was architected using a modular, event-driven approach:

1. Data Ingestion & Analysis: o Sample production sensor data is analyzed using a Lambda function built from a Python-based Docker container image. o An anomaly detection model (using scikit-learn and pandas) identifies outliers based on learned normal patterns.
2. AI Decision & Automation Layer: o Once an anomaly is detected, an AWS Lambda microservice triggers a Salesforce integration. o A secure connection is made via AWS Secrets Manager, where Salesforce credentials are stored and retrieved programmatically.
3. Salesforce Integration: o A Case Creator Lambda Function automatically generates a Service Cloud Case with relevant telemetry and issue details. o This ensures immediate visibility for service teams and enables predictive maintenance tracking.
4. Deployment & Scalability: o The full stack is orchestrated and deployed via AWS SAM (Serverless Application Model). o To handle dependency size limits, the core ML function was containerized using the AWS Lambda Python base image and published to Amazon ECR.

Challenges we ran into

• Lambda Deployment Limits: The model dependencies exceeded 250 MB, requiring a switch from ZIP-based to ECR container-based deployment. • Circular Dependencies in SAM: Adjusting the API Gateway and Lambda linkage was crucial to resolve deployment rollbacks. • OCI Image Compatibility: Rebuilding the image using an AWS Lambda base image ensured ECR compatibility and successful deployment. • Salesforce Authentication: Managing secure OAuth-based access through Secrets Manager required fine-tuning IAM policies.

Accomplishments that we're proud of

The result is a fully automated, cloud-native AI agent that can: • Detect manufacturing anomalies in real time. • Automatically create Salesforce cases for quick resolution. • Demonstrate an end-to-end intelligent automation pipeline using AWS and Salesforce.

What we learned

This project deepened my expertise in: • Building and deploying container-based AWS Lambda functions. • Optimizing serverless ML workloads under deployment constraints (ZIP vs Image). • Secure Salesforce integration via AWS Secrets Manager. • Automating anomaly-driven workflows with real-time data. • Using Infrastructure as Code (IaC) with AWS SAM and CloudFormation.

What's next for Autonomous AI Manufacturing Agent

Future improvements include: • Integrating Amazon SageMaker for more advanced anomaly detection models. • Expanding Salesforce integration to trigger Field Service workflows. • Adding an Experience Cloud dashboard for monitoring AI-driven factory insights.

Built With

• amazon-web-services
• and
• api
• between
• configuration
• data-preprocessing
• for
• functions.
• json
• lambda
• programming-languages-?-python-3.10-?-core-language-for-machine-learning-logic
• requests/responses
• sam
• template
• template.yaml).
• used
• yaml