Project Summary

Hands-on Udacity project where I set up AWS infrastructure and implemented a complete Generative AI workflow: model evaluation, fine-tuning, and endpoint deployment using AWS services. Two parallel builds (Project-1 and Project-2) validate repeatability across datasets, including provisioning on EC2, using SageMaker for training/inference, and managing datasets/artifacts on S3. The project emphasizes production hygiene like cost controls and endpoint cleanup.

Skills Demonstrated

Tools Used

Solution

Provisioned compute on EC2 and orchestrated SageMaker jobs to evaluate baseline model performance, fine-tune with domain datasets, and deploy real-time endpoints. Managed datasets and artifacts on S3, enforced IAM least-privilege access, and automated cleanup of endpoints and resources to control costs. Both project variants (IT and Finance datasets) reproduce the pipeline to confirm robustness.

Approach

1. Environment: Launch EC2, configure IAM roles/policies, install AWS CLI, Boto3, and SDKs.
2. Data: Prepare datasets, upload to S3, version artifacts for repeatability.
3. Evaluation: Run baseline inference/evaluation notebooks against chosen foundation models.
4. Fine-tuning: Train with SageMaker, track metrics, iterate on hyperparameters.
5. Deployment: Create and test inference endpoints, validate latency and outputs.
6. Ops & Cost: Monitor usage, enforce tagging, and delete endpoints/buckets when done.

Project Link(s)